Full download Artificial Intelligence and Deep Learning in Pathology 1st Edition Stanley Cohen Md (Editor) file pdf all chapter on 2024

Artificial Intelligence and Deep
Learning in Pathology 1st Edition

Stanley Cohen Md (Editor)
Visit to download the full and correct content document:
https://ebookmass.com/product/artificial-intelligence-and-deep-learning-in-pathology-
1st-edition-stanley-cohen-md-editor/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Risk Modeling: Practical Applications of Artificial

Intelligence, Machine Learning, and Deep Learning
Terisa Roberts
https://ebookmass.com/product/risk-modeling-practical-
applications-of-artificial-intelligence-machine-learning-and-
deep-learning-terisa-roberts/
Pro Deep Learning with TensorFlow 2.0: A Mathematical

Approach to Advanced Artificial Intelligence in Python
2nd Edition Santanu Pattanayak
https://ebookmass.com/product/pro-deep-learning-with-
tensorflow-2-0-a-mathematical-approach-to-advanced-artificial-
intelligence-in-python-2nd-edition-santanu-pattanayak/
Artificial Intelligence and Machine Learning for EDGE

Computing 1st Edition Rajiv Pandey
https://ebookmass.com/product/artificial-intelligence-and-
machine-learning-for-edge-computing-1st-edition-rajiv-pandey/
Artificial Intelligence and Machine Learning in the

Travel Industry: Simplifying Complex Decision Making
Ben Vinod
machine-learning-in-the-travel-industry-simplifying-complex-
decision-making-ben-vinod/
Sustainable Developments by Artificial Intelligence and
Machine Learning for Renewable Energies Krishna Kumar
https://ebookmass.com/product/sustainable-developments-by-
artificial-intelligence-and-machine-learning-for-renewable-
energies-krishna-kumar/
Artificial Intelligence and International Relations

Theories 1st Edition Bhaso Ndzendze
international-relations-theories-1st-edition-bhaso-ndzendze/
Predicting Heart Failure : Invasive, Non-Invasive,

Machine Learning, and Artificial Intelligence Based
Methods 1st Edition Kishor Kumar Sadasivuni
https://ebookmass.com/product/predicting-heart-failure-invasive-
non-invasive-machine-learning-and-artificial-intelligence-based-
methods-1st-edition-kishor-kumar-sadasivuni/
Visualization Techniques for Climate Change with

Machine Learning and Artificial Intelligence Arun Lal
Srivastav
https://ebookmass.com/product/visualization-techniques-for-
climate-change-with-machine-learning-and-artificial-intelligence-
arun-lal-srivastav/
Artificial Intelligence for Dummies 2nd Edition John

Paul Mueller & Luca Massaron
https://ebookmass.com/product/artificial-intelligence-for-
dummies-2nd-edition-john-paul-mueller-luca-massaron/
Artificial Intelligence
and Deep Learning in
Pathology
Edited by
Stanley Cohen, M.D

Emeritus Chair of Pathology & Emeritus Founding Director
Center for Biophysical Pathology
Rutgers-New Jersey Medical School, Newark, NJ, United States
Adjunct Professor of Pathology
Perelman Medical School
University of Pennsylvania, Philadelphia, PA, United States
Kimmel School of Medicine
Jefferson University, Philadelphia, PA, United States
Elsevier
Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
Copyright © 2021 Elsevier Inc. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any
means, electronic or mechanical, including photocopying, recording, or any information
storage and retrieval system, without permission in writing from the publisher. Details on
how to seek permission, further information about the Publisher’s permissions policies
and our arrangements with organizations such as the Copyright Clearance Center and the
Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright
by the Publisher (other than as may be noted herein).
Notices
Knowledge and best practice in this ﬁeld are constantly changing. As new research and
experience broaden our understanding, changes in research methods, professional
practices, or medical treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in
evaluating and using any information, methods, compounds, or experiments described
herein. In using such information or methods they should be mindful of their own safety
and the safety of others, including parties for whom they have a professional
responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or
editors, assume any liability for any injury and/or damage to persons or property as a
matter of products liability, negligence or otherwise, or from any use or operation of any
methods, products, instructions, or ideas contained in the material herein.
Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library
ISBN: 978-0-323-67538-3
For information on all Elsevier publications visit our website at

https://www.elsevier.com/books-and-journals
Publisher: Dolores Meloni

Acquisitions Editor: Michael Houston
Editorial Project Manager: Sam W. Young
Project Manager: Niranjan Bhaskaran
Cover Designer: Alan Studholme
Typeset by TNQ Technologies
This book is dedicated to the American Society for Investigative
Pathology, and especially Mark Sobel and Martha Furie, for
supporting my efforts to proselytize on behalf of artificial intelligence
in pathology. It is also dedicated to the members of the Warren Alpert
Center for Digital and Computational Pathology at Memorial Sloan-
Kettering, under the leadership of David Klimstra and Meera
Hameed, for nourishing my enthusiasm for this revolution in
pathology. I am especially indebted to both Dean Christine Palus and
the Departments of Computer Science and Mathematics at Villanova
University for enabling me to get up to speed rapidly in this relatively
new interest of mine while being so accommodating and friendly to a
student older than his teachers. Most important, it is also dedicated
to my wonderful wife and scientific colleague Marion, children
(Laurie, Ron, and Ken), kids-in-law (Ron, Helen, and Kim
respectively), and grandkids (Jessica, Rachel, Joanna, Julie, Emi,
Risa, Brian, Seiji, and Ava). With natural intelligence like that, who
needs artificial intelligence?
Contributors
Tanishq Abraham
Department of Biomedical Engineering, University of California, Davis, CA,
United States
Ognjen Arandjelovi
c, M.Eng. (Oxon), Ph.D. (Cantab)
School of Computer Science, University of St Andrews, St Andrews, United
Kingdom
Peter D. Caie, BSc, MRes, PhD
School of Medicine, QUAD Pathology, University of St Andrews, St Andrews,
United Kingdom
Stanley Cohen, MD
Emeritus Chair of Pathology & Emeritus Founding Director, Center for Biophysical
Pathology, Rutgers-New Jersey Medical School, Newark, NJ, United States;
Adjunct Professor of Pathology, Perelman Medical School, University of
Pennsylvania, Philadelphia, PA, United States; Adjunct Professor of Pathology,
Kimmel School of Medicine, Jefferson University, Philadelphia, PA, United States
Neofytos Dimitriou, B.Sc
School of Computer Science, University of St Andrews, St Andrews, United
Kingdom
Michael Donovan
Department of Pathology, Icahn School of Medicine at Mount Sinai, New York,
NY, United States
Gerardo Fernandez
NY, United States
Rajarsi Gupta, MD, PhD
Assistant Professor, Biomedical Informatics, Stony Brook Medicine, Stony Brook,
NY, United States
Richard Levenson, MD
Professor and Vice Chair, Pathology and Laboratory Medicine, UC Davis Health,
Sacramento, CA, United States
Bahram Marami
NY, United States
Benjamin R. Mitchell, BA, MSE, PhD
Assistant Professor, Department of Computing Sciences, Villanova University,
Villanova, PA, United States
xiii
xiv Contributors
Daniel A. Orringer, MD
Associate Professor, Department of Surgery, NYU Langone Health, New York City,
NY, United States
Anil V. Parwani, MD, PhD, MBA
Professor, Pathology and Biomedical Informatics, The Ohio State University,
Columbus, OH, United States
Marcel Prastawa
NY, United States
Abishek Sainath Madduri
NY, United States
Joel Haskin Saltz, MD, PhD
Professor, Chair, Biomedical Informatics, Stony Brook University, Stony Brook,
NY, United States
Richard Scott
NY, United States
Austin Todd
UT Health, San Antonio, TX, United States
John E. Tomaszewski, MD
SUNY Distinguished Professor and Chair, Pathology and Anatomical Sciences,
University at Buffalo, State University of New York, Buffalo, NY, United States
Jack Zeineh
NY, United States
Preface
It is now well-recognized that artificial intelligence (AI) represents an inflection

point for society at least as profound as was the Industrial Revolution. It plays
increasing roles in industry, education, recreation, and transportation, as well as sci-
ence and engineering. As one example of its ubiquity, a recent headline proclaimed
that “Novak Djokovic used AI to Train for Wimbledon!”
In particular, AI has become important for both biomedical research and clin-
ical practice. Part of the reason for this is the sheer amount of data currently
obtainable by modern techniques. In the past, we would study one or a few en-
zymes or transduction pathways at a time. Now we can sequence the entire
genome, develop gene expression arrays that allow us to dissect out critical net-
works and pathways, study the entire microbiome instead of a few bacterial spe-
cies in isolation, and so on. We can also use AI to look for important patterns in
the massive data that have already been collected, i.e., data mining both in defined
datasets and in natural language sources. Also, we can take advantage of the power
of AI in pattern recognition to analyze and classify pictorial data not only for clas-
sification tasks but also to extract information not obvious to an individual limited
to direct observations through the microscope, a device that has never received
FDA approval.
Radiologists have been among the first clinical specialists to embrace AI to assist
in the processing and interpretation of X-ray images. Pathologists have lagged, in
part because the visual content of the images that pathologists use is at least an order
of magnitude greater than those of radiology. Another factor has been the delay of
the major industrial players to enter this arena, as well as limited hospital resources
based on perceived differences in hospital profitability between radiology and pa-
thology practices. However, more and more pathology departments are embracing
digital pathology via whole slide imaging (WSI), and WSI provides the infrastruc-
ture both for advanced imaging modalities and the archiving of annotated datasets
needed for AI training and implementation. An ever-increasing number of articles
utilizing AI are appearing in the literature, and a number of professional societies
devoted to this field have emerged. Existing societies and journals in pathology
have begun to create subsections specifically devoted to AI.
Unfortunately, existing texts on machine learning, deep learning, and AI (terms
that are distinct, but loosely overlap) tend to be either very general and superficial
overviews for the layman or highly technical tomes assuming a background in linear
algebra, multivariate calculus, and programming languages. They do not therefore
serve as a useful introduction for most pathologists, as well as physicians in other
xv
xvi Preface
disciplines. The purpose of this monograph is to provide a detailed understanding of

both the concepts and methods of AI, as well as its use in pathology. In broad general
terms, the book covers machine learning and introduction to data processing, the
integration of AI and pathology, and detailed examples of state-of-the-art AI appli-
cations in pathology. It focuses heavily on diagnostic anatomic pathology, and
thanks to advances in neural network architecture we now have the tools for extrac-
tion of high-level information from images.
We begin with an opening chapter on the evolution of machine learning,
including both a historical perspective and some thoughts as to its future. The
next four chapters provide the underlying framework of machine learning with spe-
cial attention to deep learning and AI, building from the simple to the complex.
Although quite detailed and intensive, these require no knowledge of computer pro-
gramming or of mathematical formalism. Wherever possible, concepts are illus-
trated by examples from pathology. The next two chapters deal with the
integration of AI with digital pathology and its application not only for classifica-
tion, prediction, and inference but also for the manipulation of the images them-
selves for extraction of meaningful information not readily accessible to the
unaided eye. Following this is a chapter on precision medicine in digital pathology
that demonstrates the power of traditional machine learning paradigms as well as
deep learning, The reader will find a certain amount of overlap in these chapters
for the following reasons; each author needs to review some basic aspects of AI
as they specifically relate to his or her assigned topic, each presents their own
perspective on the underpinnings of the work they describe, and each topic consid-
ered can best be understood in the light of parallel work in other areas. The next two
chapters deal with two important topics relating to cancer in which the authors uti-
lize AI to great effect and discuss their findings in the context of studies in other lab-
oratories. Taken together, these latter chapters provide complementary views of the
current state of the art based upon different perspectives and priorities.
The book concludes with an overview that puts all this in perspective and makes
the argument that AI does not replace us, but rather serves as a digital assistant.
While it is still too early to determine the final direction in which pathology will
evolve, it is not too early to recognize that it will grow well beyond its current ca-
pabilities and will do so with the aid of AI. However, until “general AI” is attained
that can fully simulate conscious thought (which may never happen), biological
brains will remain indispensable. Thus, for the foreseeable future, human patholo-
gists will not be obsolete. Instead, the partnership between artificial and natural in-
telligence will continue to expand to form a true partnership. The machine will
Preface xvii
generate knowledge, and the human will provide the wisdom to properly apply that
knowledge. The machine will provide the knowledge that a tomato is a fruit. The
human will provide the wisdom that it does not belong in a fruit salad. Together
they will determine that ketchup is not a smoothie.
Stanley Cohen, MD
Emeritus Chair of Pathology & Emeritus Founding Director
Center for Biophysical Pathology
Rutgers-New Jersey Medical School, Newark, NJ, United States
Perelman Medical School
University of Pennsylvania, Philadelphia, PA, United States
Kimmel School of Medicine
Jefferson University, Philadelphia, PA, United States
CHAPTER
The evolution of machine

learning: past, present,
and future 1
Stanley Cohen, MD 1, 2, 3
1
Emeritus Chair of Pathology & Emeritus Founding Director, Center for Biophysical Pathology,
Rutgers-New Jersey Medical School, Newark, NJ, United States; 2Adjunct Professor of Pathology,
Perelman Medical School, University of Pennsylvania, Philadelphia, PA, United States; 3Adjunct
Professor of Pathology, Kimmel School of Medicine, Jefferson University, Philadelphia, PA, United
States
Introduction
The first computers were designed solely to perform complex calculations rapidly.
Although conceived in the 1800s, and theoretical underpinnings developed in the
early 1900s, it was not until ENIAC was built that the first practical computer can
be said to have come into existence. The acronym ENIAC stands for Electronic
Numerical Integrator and Calculator. It occupied a 20 40 foot room, and contained
over 18,000 vacuum tubes. From that time until the mid-20th century, the programs
that ran on the descendants of ENIAC were rule based, in that the machine was given
a set of instructions to manipulate data based upon mathematical, logical, and/or
probabilistic formulas. The difference between an electronic calculator and a
computer is that the computer encodes not only numeric data but also the rules
for the manipulation of these data (encoded as number sequences).
There are three basic parts to a computer: core memory, where programs and data
are stored, a central processing unit that executes the instructions and returns the
output of that computation to memory for further use, and devices for input and
output of data. High-level programming languages take as input program statements
and translate them into the (numeric) information that a computer can understand.
Additionally, there are three basic concepts that are intrinsic to every programming
language: assignment, conditionals, and loops. In a computer, x ¼ 5 is not a state-
ment about equality. Instead, it means that the value 5 is assigned to x. In this
context, X ¼ X þ 2 though mathematically impossible makes perfect sense. We
add 2 to whatever is in X (in this case, 5) and the new value (7) now replaces that
old value (5) in X. It helps to think of the ¼ sign as meaning that the quantity on
the left is replaced by the quantity on the right. A conditional is easier to understand.
We ask the computer to make a test that has two possible results. If one result holds,
then the computer will do something. If not, it will do something else. This is typi-
cally an IF-THEN statement. IF you are wearing a green suit with yellow polka dots,
Artificial Intelligence and Deep Learning in Pathology. https://doi.org/10.1016/B978-0-323-67538-3.00001-4 1

Copyright © 2021 Elsevier Inc. All rights reserved.
2 CHAPTER 1 The evolution of machine learning: past, present, and future
PRINT “You have terrible taste!“ IF NOT print “There is still hope for your sartorial
sensibilities.” Finally, a LOOP enables the computer to perform a single instruction
or set of instructions a given number of times. From these simple building blocks, we
can create instructions to do extremely complex computational tasks on any input.
The traditional approach to computing involves encoding a model of the problem
based on a mathematical structure, logical inference, and/or known relationships
within the data structure. In a sense, the computer is testing your hypothesis with
a limited set of data but is not using that data to inform or modify the algorithm itself.
Correct results provide a test of that computer model, and it can then be used to work
on larger sets of more complex data. This rules-based approach is analogous to
hypothesis testing in science. In contrast, machine learning is analogous to
hypothesis-generating science. In essence, the machine is given a large amount of
data and then models its internal state so that it becomes increasingly accurate in
its predictions about the data. The computer is creating its own set of rules from
data, rather than being given ad hoc rules by the programmer. In brief, machine
learning uses standard computer architecture to construct a set of instructions that,
instead of doing direct computation on some inputted data according to a set of rules,
uses a large number of known examples to deduce the desired output when an
unknown input is presented. That kind of high-level set of instructions is used to
create the various kinds of machine learning algorithms that are in general use today.
Most machine learning programs are written in PYTHON programming language,
which is both powerful and one of the easiest computer languages to learn.
A good introduction and tutorial that utilizes biological examples is Ref. [1].
Rules-based versus machine learning: a deeper look

Consider the problem of distinguishing squares and rectangles and circles from each
other. One way to approach this problem using a rules-based program would be to
calculate relationships between area and circumference for each shape by using
geometric-based formulas, and then comparing the measurements of a novel
unknown sample to those values for the best fit, which will define its shape. One
could also simply count edges, defining a circle as having zero edges, and use
that information to correctly classify the unknown.
The only hitch here would be the necessity to know pi for the case of the circle.
But that is also easy to obtain. Archimedes did this somewhere about 250 BCE by
using geometric formula to calculate the areas of two regular polygons; one polygon
inscribed inside the circle and the other the polygon with which the circle was
circumscribed (Fig. 1.1).
Since the actual area of the circle lies between the areas of the inscribed and
circumscribed polygons, the calculable areas of these polygons set upper and layer
bounds for the area of the circle. Without a computer, Archimedes showed that pi is
between 3.1408 and 3.1428. Using computers, we can now do this calculation for
polygons with very large numbers of sides and reach an approximate value of pi
Rules-based versus machine learning: a deeper look 3
FIGURE 1.1
Circle bounded by inscribed and circumscribed polygons. The area of the circle lies
between the areas of these bounds. As the number of polygon sides increases, the
approximation becomes better and better.
Wikipedia commons; Public domain.
to how many decimal places we require. For a more “modern” approach, we can use
a computer to sum a very large number of terms of an infinite series that converges to
pi. There are many such series, the first of which having been described in the 14th
century.
These rules-based programs are also capable of simulations; as an example,
pi can be computed by simulation. For example, if we have a circle quadrant of
radius ¼ 1 embedded in a square target board with sides ¼ 1 units, and throw darts
at the board so that they randomly land somewhere within the board, the probability
of a hit within the circle segment is the ratio of the area of that segment to the area of
the whole square, which by simple geometry is pi/4. We can simulate this experi-
ment by computer, by having it pick pairs of random numbers (each between zero
and one) and using these as X-Y coordinates to compute the location of a simulated
hit. The computer can keep track of both the ratio of the number of hits landing
within the circle quadrant and the total number of hits anywhere on the board,
and this ratio is then equal to pi/4. This is illustrated in Fig. 1.2.
FIGURE 1.2
Simulation of random dart toss. Here the number of tosses within the segment of the
circle is 5 and the total number of tosses is 6.
From Wikipedia commons; public domain.
In the example shown in Fig. 1.2, the value of pi is approximately 3.3, based on
only 6 tosses. With only a few more tosses we can beat Archimedes, and with a very
large number of tosses we can come close to the best value of pi obtained by other
means. Because this involves random processes, it is known as Monte Carlo
simulation.
Machine learning is quite different. Consider a program that is created to distin-
guish between squares and circles. For machine learning, all we have to do is present
a large number of known squares and circles, each labeled with its correct identity.
From this labeled data the computer creates its own set of rules, which are not given
to it by the programmer. The machine uses its accumulated experience to ultimately
correctly classify an unknown image. There is no need for the programmer to explic-
itly define the parameters that make the shape into a circle or square. There are many
different kinds of machine learning strategies that can do this.
In the most sophisticated forms of machine learning such as neural networks
(deep learning), the internal representations by which the computer arrives at a clas-
sification are not directly observable or understandable by the programmer. As the
program gets more and more labeled samples, it gets better and better at distinguish-
ing the “roundness,” or “squareness” of new data that it has never seen before. It is
using prior experience to improve its discriminative ability, and the data itself need
not be tightly constrained. In this example, the program can arrive at a decision that
is independent of the size, color, and even regularity of the shape. We call this
artificial intelligence (AI), since it seems more analogous to the way human beings
behave than traditional computing.
Varieties of machine learning

The age of machine learning essentially began with the work of pioneers such as
Marvin Minsky and Frank Rosenblatt who described the earliest neural networks.
However, this approach languished for a number of reasons including the lack of
algorithms that could generalize to most real-world problems, the primitive nature
of input storage, and computation in that period. The major conceptual innovations
that moved the field forward were backpropagation and convolution, which we will
discuss in detail in a later chapter. Meanwhile, other learning algorithms that were
being developed also went beyond simply rule-based calculation and involved infer-
ences from large amounts of data. These all address the problem of classifying an
unknown instance based upon the previous training of the algorithm with a known
dataset.
Most, but not all, algorithms fall into three categories. Classification can be based
on clustering of data points in an abstract multidimensional space (support vector
machines), probabilistic considerations (Bayesian algorithms), or stratification of
the data through a sequential search in decision trees and random forests. Some
methods utilizing these techniques are shown in Fig. 1.3.
Varieties of machine learning 5
FIGURE 1.3
Varieties of machine learning.
Reprinted from https://medium.com/sciforce/top-ai-algorithms-for-healthcare-aa5007ffa330, with permission
of Dr. Max Ved, Co-founder, SciForce.
All of those latter approaches, in addition to neural networks, are collectively

known as “machine learning.” In recent years, however, neural networks have
come to dominate the field because of their near-universal applicability and their
ability create high-level abstractions from the input data. Neural network-based
programs are often defined as “deep learning.” Contrary to common belief, this is
not because of the complexity of the learning that takes place in such a network,
but simply due to the multilayer construction of a neural network as will be dis-
cussed below. For simplicity, in this book we will refer to the other kinds of machine
learning as “shallow learning.”
The evolution of the neural network class of machine learning algorithms has
been broadly but crudely based upon the workings of biological neural networks
in the brain. The neural network does not “think”; rather it utilizes information to
improve its performance. This is learning, rather than thinking. The basics of all
of these will be covered in this chapter. The term “Artificial Intelligence” is often
used to describe the ability of a computer to perform what appears to be human-
level cognition, but this is an anthropomorphism of the machine, since, as stated
above, machines do not think. In the sense that they learn by creating generalizations
from inputted training data, and also because certain computer programs can be
trained by a process similar to that of operant conditioning of an animal, it is easy
to think of them as intelligent. However, as we shall see, machine learning at its
most basic level is simply about looking for similarities and patterns in things.
Our programs can create high levels of abstraction from building blocks such as
these in order to make predictions, and identify patterns. But “asking if a machine
can think is like asking whether a submarine can swim” [2]. Nevertheless, we will
use the term artificial intelligence throughout the book (and in the title as well) as
it is now firmly ingrained in the field.
General aspects of machine learning

The different machine learning approaches for chess provide good examples of the
differences between rules-based models and AI models. To program a computer to
play chess, we can give it the rules of the game, some basic strategic information
along the lines of “it’s good to occupy the center,” and data regarding typical chess
openings. We can also define rules for evaluating every position on the board for
every move. The machine can then calculate all possible moves from a given posi-
tion and select those that lead to some advantage. It can also calculate all possible
responses by the opponent. This brute force approach evaluated billions of possible
moves, but it was successful in defeating expert chess players. Machine learning
algorithms, on the other hand, can develop strategically best lines of play based
upon the experiences of learning from many other games played previously, much
as prior examples improved a simple classification problem as described in the
previous section. Early machine learning models took only 72 h of play to match
the best rules-based chess programs in the world. Networks such as these learn to
make judgment without implicit input as to the rules to be invoked in making those
judgments.
What most machine learning programs have in common is the need for large
amounts of known examples as the data with which to train them. In an abstract
sense, the program uses data to create an internal model or representation that encap-
sulates the properties of those data. When it receives an unknown sample, it deter-
mines whether it fits that model to “train” them. A neural network is one of the best
approaches toward generating such internal models, but in a certain sense, every
machine learning algorithm does this. For example, a program might be designed
to distinguish between cats and dogs. The training set consists of many examples
of dogs and cats. Essentially, the program starts by making guesses, and the guesses
are refined through multiple known examples. At some point, the program can clas-
sify an unknown example correctly as a dog or a cat. It has generated an internal
model of “cat-ness” and “dog-ness.” The dataset composed of known examples is
known as the “training set.” Before using this model on unlabeled, unknown exam-
ples, one must test the program to see how well it works. For this, one also needs a
set of known examples that are NOT used for training, but rather to test whether the
machine has learned to identify examples that it has never seen before. The algo-
rithm is analyzed for its ability to characterize this “test set.”
Although many papers quantitate the results of this test in terms of accuracy, this
is an incomplete measure, and in cases where there is an imbalance between the two
classes that are to be distinguished, it is completely useless. As an example, overall
cancer incidence for the period 2011e15 was less than 500 cases per 100,000 men
and women. I can create a simple algorithm for predicting whether a random indi-
vidual has cancer or not, by simply having the algorithm report that everyone is
negative. This model will be wrong less than only 500 times, and so its accuracy
is approximately 99.5%! Yet this model is clearly useless, as it will not identify
any patient with cancer.
Deep learning and neural networks 7
As this example shows, we must not only take into account the accuracy of the
model but also the nature of its errors. There are several ways of doing this. The
simplest is as follows. There are four possible results for an output from the model:
true positives (TP), false positives (FP), true negatives (TN), and false negatives
(FN). There are simple formulas to determine the positive predictive value and
the sensitivity of the algorithm. These are known as “precision” (Pr) and “recall”
(Re), respectively. In some publications, an F1 value is given, which is the harmonic
mean of Pr and Re.
Deep learning and neural networks

For many kinds of machine learning, the programmer determines the relevant
features and their quantification. Features may include such things as weight, size,
color, temperature, or any parameter that can be described in quantitative, semiquan-
titative, or relative terms. The process of creating feature vectors is often known as
feature engineering. There are some things that have too many features to do this
directly. For example, a picture may have several megabytes of data at the pixel
level. Prior to machine learning, image analysis was based upon specific rules and
measurements to define features, and a rules-based program detected patterns based
on edges, textures, shapes, and intensity to identify objects of such as nuclei, cyto-
plasm, mitoses, necrosis and so on, but interpretation of these was up to the human
pathologist. Before neural networks, machine learning also needed humans to define
features, a task often referred to “feature engineering.” These more powerful net-
works could make specific classification and predictions from the image as a whole.
Neural networks consisting of layers of simulated neurons took this a step further; a
large part of the strength of these algorithms is that the network can abstract out
high-level features from the inputted data all by itself. Those are based on combina-
tions of initial data inputs and are often referred to as representations of the data.
These representations are of significantly lower dimensionality than the original
feature vectors of the input. The representations are formed in the network layers
that are between input and output which are referred to as hidden layers. Training
refines these representations, and the program uses them to derive its own internal
rules for decision-making based on them. Thus, the neural network learns rules
by creating hidden variables from data.
Within the inner layers of the neural network, the process of abstraction leading
to a representation is based on an iterative model often initiated by random variables,
and so is partially unexplainable to the human observer (which is why the term
“hidden” is used for these inner layers). This lack of explainability is uncomfortable
for some researchers in the field. However, if you were to ask an experienced pathol-
ogist how he or she arrived a diagnosis based upon examination of a biopsy slide,
they would find it difficult to explain the cognitive processes that went into that judg-
ment, although if pressed, they could define a set of features consistent with that
diagnosis. Thus, both neural networks and humans base their decisions on processes
not readily explainable.
The simplest neural net has at least one hidden layer. Most neural nets have
several layers of simulated neurons, and the “neurons” of each layer are connected
to all of the neurons in the next layer. These are often preceded by a convolutional set
of layers, based loosely on the architecture of the visual cortex of the brain. These
transform the input data to a form that can be handled by the following fully
connected layer. There are many different kinds of modern sophisticated machine
learning algorithms based on this kind of neural network, and they will be featured
predominantly throughout this monograph. Not all machine learning programs are
based on these deep learning concepts. There are some that rely on different inter-
connections among the simulated neurons within them, for example, so-called
restricted Boltzmann networks. Others, such as spiking neural networks, utilize an
architecture that is more like biological brains work. While these represent examples
of the current state of the art in AI, the basic fully connected and convolutional
framework are simpler to train and more computationally efficient, and thus remain
the networks of choice as of 2019.
It should also be noted that it is possible to do unsupervised or weakly supervised
learning where there is little to no a priori information about the class labels and
large datasets are required for these approaches as well. In subsequent chapters
we will describe these concepts in detail as we describe the inner workings of the
various kinds of machine learning, and following these we will describe how AI
fits into the framework of whole slide imaging (WSI) in pathology. With this as base-
line, we will explore in detail examples of current state-of-the art applications of AI
in pathology. This is especially timely in that AI is now an intrinsic part of our lives.
The modern safety features in cars, decisions on whether or not we are good credit
risks, targeted advertising, and facial recognition are some obvious examples. When
we talk to Siri or Alexa we are interacting with an AI program. If you have an iWatch
or similar device, you can be monitored for arrhythmias, accidental falls, etc. AI’s
successes have included natural language processing and image analysis, among
other breakthroughs.
AI plays an increasingly important role in medicine, from research to clinical
applications. A typical workflow for using machine learning for integrating clinical
data for patient care is illustrated in Fig. 1.4.
IBM’s Watson is a good example of the use of this paradigm. Unstructured data
from natural language is processed and outputted in a form that can be combined
with the more structured clinical information such as laboratory tests and physical
diagnosis. In the paradigm shown here, images are not themselves interpreted via
machine learning; rather they are the annotation of those entities (diagnostic impres-
sion, notable features, etc.) that are used. As neural networks have become more
sophisticated, they can take on the task of interpreting the images and, under pathol-
ogist and radiologist supervision, further streamline the diagnostic pathway.
In Fig. 1.4, AI is the term used to define the whole process by which the machine
has used its training to make what appears to be a humanlike judgment call.
The role of AI in pathology 9
FIGURE 1.4
The incorporation of artificial intelligence into clinical decision-making. Both precise
quantifiable data and unstructured narrative information can be captured within this
framework.
Reprinted from https://medium.com/sciforce/top-ai-algorithms-for-healthcare-aa5007ffa330, with permission
of Dr. Max Ved, Co-founder, SciForce.
The role of AI in pathology

It has been estimated that approximately 70% of a clinical chart is composed of lab-
oratory data, and the pathologist is therefore an important interpreter of those data in
the context of the remaining 30%. Even in the age of molecular analysis, the cogni-
tive skills involved in the interpretation of image-based data by the pathologist
remain of major importance in those clinical situations where diagnosis occurs at
the tissue level. This has become increasingly difficult because of the complexities
of diagnostic grading for prognosis, the need to integrate visual correlates of molec-
ular data such as histochemical markers based on genes, gene expression, and immu-
nologic profiles, and the wealth of archival material for comparative review. For
these and many other reasons, research and implementation of AI in anatomic pa-
thology, along with continuing implementation in clinical pathology, is proceeding
at an exponential rate.
In many ways, the experiences of radiologists are helping to guide us in the

evolution of pathology, since radiologists continue to adopt AI at a rapid pace.
Pathologists have been late in coming to the table. For radiologists AI is mainly
about analyzing images, and their images are relatively easy to digitize. In pathol-
ogy, acquisition and annotation of datasets for computer input was not possible until
WSI at appropriate scale of resolution and speed of acquisition became available.
Some of the drivers of WSI are efficiency, data portability, integration with other
diagnostic modalities, and so on, but two main consequences of the move from glass
microscopy to WSI are the ability to incorporate AI and the incorporation of
computer-intensive imaging modalities. Although AI plays important roles in
clinical pathology, in the interpretation of genomic and proteomic arrays, etc., where
inference and prediction models are important, the focus of this book will be on the
use of AI in anatomic pathology, with specific attention to the extraction of maximal
information from images regarding diagnosis, grading, prediction, and so on.
Limitations of AI
While recognizing the great success of AI, it is important to note three important
limitations. First, for the most part, current algorithms are one-trick ponies that
can perform a specific task that they have been trained for, with limited capability
of applying that training to new problems. It is possible to do so-called transfer
learning on a computer, but it is still far removed from the kind of general intelli-
gence that humans possess. Second, training an algorithm requires huge datasets
often numbering in the thousands of examples. In contrast, humans can learn
from very small amounts of data. It does not take a toddler 1000 examples to learn
the difference between an apple and a banana. Third, computation by machine is
very inefficient, in terms of energy needs as compared with the human brain,
even though the brain is much more complex than the largest supercomputer ever
developed. A typical human brain has about 100 billion neurons and 100 trillion
interconnections, which is much bigger than the entire set of neural networks
running at any time across the globe. This has a secondary consequence. It has
been estimated that training large AI models produces carbon emissions nearly
five times greater than the lifecycle carbon emissions of an automobile.
There are several new approaches that have the ability to address these issues.
The first is neuromorphic computing, which attempts to create a brainlike physical
substrate in which the AI can be implemented, instead of running solely as a soft-
ware model. Artificial neurons, based upon use of a phase-change material, have
been created. A phase-change material is one that can exist in two different phases
(for example, crystalline or amorphous) and easily switch between the two when
triggered by laser or electricity. At some point the material changes from a noncon-
ductive to a conductive state and fires.
After a brief refractory period, it is ready to begin this process again. In other
words, this phase-change neuron behaves just like a biological neuron. Artificial
synapses have also been created. These constructs appear to be much more energy
The role of AI in pathology 11
efficient than current fully software implementations such as both conventional

neural networks and spiking net algorithms that more directly simulate neuronal
behavior.
Much further in the future is the promise of quantum computing that will be
orders of magnitude faster than the most powerful computers presently available.
These are based on three-state “qubits” (0, 1, and an indeterminate superposition
of these) rather than binary bits, and these states can hold multiple values simulta-
neously. A quantum computer can also process these multiple values simultaneously.
In theory, this allows us to address problems that are currently intractable and could
lead to major advances in AI. Although current quantum computers have only a
small number of interacting qubits, it is conceivable that a quantum computer of
the future could approach the complexity of the human brain, with far less numbers
of interacting qubits (unless, of course, the brain itself functions in part as a quantum
computer, as some have speculated). This is not as far-fetched as it seems, as photo-
synthetic processes have been shown to exhibit quantum properties at physiological
temperatures [3,4], thus demonstrating the feasibility of maintaining quantum coher-
ence at room temperatures, based on technologies that have had millions of years to
evolve.
Even with current technology, however, both the potential and current applica-
tions of AI have demonstrated that it will play as powerful a role in pathology
and medicine in general, as it has already done in many other fields and disciplines.
For both a good overview and historical perspective of AI, see Ref. [5].
General aspects of AI
For any use of AI in solving real-world problems, a number of practical issues must
be addressed. For example,
a. For the choice of a given AI algorithm, what are the classes of learning problems
that are difficult or easy, as compared with other potential strategies for their
solution?
b. How does one estimate the number of training examples necessary or sufficient
for learning?
c. What trade-offs between precision and sensitivity is one willing to make?
d. How does computational complexity of the proposed model compare with that of
competing models?
For questions such as these there is no current analytical approach to the best
answer. Thus, the choice of algorithm and its optimization remains an empirical trial
and error approach. Nevertheless, there are many aspects of medicine that have
proven amenable to AI approaches. Some are “record-centric” in that they mine
electronic health records, journal and textbook materials, patient clinical support
systems, and so on, for epidemiologic patterns, risk factors, and diagnostic predic-
tions. These all involve natural language processing. Others involve signal classifi-
cation, such as interpretation of EKG and EEG recordings. A major area of
Another random document with
no related content on Scribd:
of the intellectual life there, this biological influence is felt more
acutely in America than elsewhere.
The one outstanding characteristic again of every approved
academic teacher in the United States is his method of handling his
subject, just as the one thing that is claimed for Pragmatism by its
upholders is that it is particularly a “methodology” of thought rather
than a complete philosophy. To the university constituency of the
United States a professor without an approved and successful
method is as good as dead, for no one would listen to him. The most
manifest sign, to be sure, of the possession of such an effective
method on the part of the university lecturer is the demonstration of
skill in the treatment of his subject, in the “approach” that he makes
to it for the beginner, in his power of setting the advanced student to
work upon fruitful problems, and of giving him a complete
“orientation” in the entire field under consideration. And then in
addition to this he must be able to indicate the practical and the
educational value of what he is teaching.
In his review of James’s classical work upon Pragmatism,
Dewey, while indicating a number of debatable points in the
pragmatist philosophy, declares emphatically his belief in that
philosophy as a method of “orientation.” The title again of Peirce’s
famous pamphlet was How to make Ideas Clear—a phrase of itself
suggestive enough of the inquiring mind of the young student when
oppressed by apparently conflicting and competing points of view.
“We are acquainted with a thing,” says James, “as soon as we have
learned how to behave towards it, or how to meet the behaviour we
accept from it.” In one of his books he talks about physics, for
example, as giving us not so much a theory about things as a
“practical acquaintance” with bodies; “the power to take hold of them
and handle them,” indicating at the same time his opinion that this
way of regarding knowledge should be extended to philosophy itself.
All of this will serve as a proof or illustration of the essentially
“practical” and “methodological” conception of philosophy taken by
the pragmatists. Papini refers, we remember, to the pragmatist
philosophy as a power of “commanding our material,” of
“manipulating” for practical purposes the different “thought-
constructions” of the history of philosophy. And those who have any
familiarity with the early pragmatist magazine literature know that the
pragmatists used to be fond of asking themselves such preliminary
and “laboratory-like” inquiries as the following: “What is truth known
as?” “What is philosophy known as?” “What are the different
‘thought-levels’ upon which we seem to move in our ordinary
experience?” They never exactly seem to “define” philosophy for
you, preferring to indicate what it can do for you, and so on.
Turning now to the matter of American characteristics that are
broader and deeper than the merely academic, we may find an
illustration, for example, of the American practicality and the love of
the concrete (instead of the abstract or the merely general) in the
following declaration of Professor James that “the whole originality of
Pragmatism, the whole point in it, is its use of the concrete way of
seeing. It begins with concreteness and returns and ends with it.” Of
the American love of novelty and of interest we may find an
illustration in the determination of Pragmatism “never to discuss a
question that has absolutely no interest and no meaning to any one.”
Of Pragmatism as an exemplification of the American love of
experiment, and of experimentation, with a view to definite and
appreciable “returns,” we may give the following: “If you fully believe
the pragmatic method you cannot look on any such word, i.e. ‘God,’
‘Matter,’ ‘Reason,’ ‘The Absolute,’ ‘Energy,’ and such ‘solving’
names, as closing your quest. You must bring out in each word its
practical cash value, set it at work within the stream of your
experience. It appears less as a solution then than as a programme
for more work and more particularly as an indication of the ways in
which existing realities may be changed.” Of the American
intolerance for mere scholarship and book-learning, and of the
American inability to leave any discovery or any finished product
alone without some attempt to “improve” upon it or to put it to some
new use, we may cite the following: “When may a truth go into cold
storage in the encyclopaedias, and when shall it come out for
battle?”
Another very strongly marked characteristic of American life is
the thoroughly eclectic and composite character of its general culture
and of the general tone of its public life. American daily life has
become, as it were, a kind of social solvent, a huge melting-pot for
the culture and the habits and the customs of peoples from all over
the earth. This also may be thought of as reflected in the confessedly
complex and amalgam-like character of Pragmatism, in its boast and
profession of being a synthesis and a fusion of so many different
tendencies of human thought. As a juxtaposition, or kind of
compound solution, of such a variety of things as the affirmations of
religion, the hypothetical method of science, realism, romanticism,
idealism, utilitarianism, and so on, it reminds us only too forcibly of
the endless number of social groups and traditions, the endless
number of interests and activities and projects to be seen and felt in
any large American city.
Still another general characteristic of American life of which we
may well think in connexion with Pragmatism is the sociological
interest of the country, the pressure of which upon the pragmatists
and their writings has already been referred to. The social problem in
295
America has now become the one problem that is present with
everybody, and present most of all, perhaps, with the European
immigrant, who has for various reasons hoped that he had left this
problem behind him. The effect of this upon Pragmatism is to be
seen, not merely in the very living hold that it is inclined to take of
296
philosophy and philosophical problems, but in the fact of its boast
of being a “way of living” as well as a “way of thinking.” We have
examined this idea in our remarks upon the ethics of Pragmatism.
Of course the outstanding temperamental American
characteristic that is most clearly seen in Pragmatism is the great
fact of the inevitable bent of the American mind to action and to
accomplishment,—its positive inability to entertain any idea, or any
set of ideas upon any subject whatsoever, without experiencing at
the same time the inclination to use these ideas for invention and
297
contrivance, for organization and exploitation. Any one who has
lived in the United States must in fact have become so habituated
and so accustomed to think of his thought and his knowledge and his
capacities in terms of their possible social utility, that he simply
cannot refrain from judging of any scheme of thought or of any set of
ideas in the same light. Anywhere, to be sure, in the United States
will they allow a man to think all he pleases about anything
whatsoever—even pre-Socratic philosophy, say, or esoteric
Buddhism. And there is nothing indeed of which the country is said,
by those who know it best, to stand so much in need as the most
persistent and the most profound thought about all important
matters. But such thought, it is always added, must prove to be
constructive and positive in character, to be directed not merely to
the solution of useless questions or of questions which have long
ago been settled by others.
298
We shall now endeavour to think of the value to philosophy
and to the thought and practice of the world (the two things are
inseparable) of some or all of these general and special
characteristics which we have sought to illustrate in Pragmatism.
We might begin by suggesting the importance to the world of the
299
production and development of a man of genius like James,
whose fresh and living presentation of the problems of philosophy
(as seen by a psychologist) has brought the sense of a lasting and
far-reaching obligation upon his fellow-students everywhere. In no
more favourable soil could James have grown up into the range and
plenitude of his influence than in that of America and of Harvard
300
University, that great nursing-ground of the finest kind of
American imperialism. The great thing, of course, about James was
301
his invasion, through the activities of his own personality, of the
realm of philosophical rationalism by the fact and the principle of
active personality. His whole general activity was a living
embodiment of the principle of all humanism, that personality and the
various phases of personal experience are of more importance to
philosophy in the way of theory than any number of supposedly self-
coherent, rational or abstract systems, than any amount of reasoning
that is determined solely by the ideal of conceptual consistency.
Then again, it might be held that the entire academic world of to-
day has a great deal to learn from the conditions under which all
subjects (philosophy included) are taught and investigated in the
typical American university of the day. We have referred to the fact
that the American professor or investigator faces the work of
instruction and research in an environment replete with all modern
302
facilities and conveniences. The very existence of this
environment along with the presence throughout his country of
university men and workers from all over the world with all their
obvious merits and defects as “social types” prevent him in a
hundred ways from that slavery to some one school of thought, to
some one method of research that is so often a characteristic of the
scholar of the old world. The entire information and scholarship in
any one science (say, philosophy) is worth to him what he can make
of it, here and now, for himself and for his age and for his immediate
environment. He simply cannot think of any idea or any line of
reflection, in his own or in any other field, without thinking at the
same time of its “consequences,” immediate, secondary, and
remote. This inability is an instance of the working of the pragmatist
element in scholarship and in thought with all its advantages and
303
disadvantages.
And it is true too, it might be held, even upon the principles of
Idealism that the mere facts of knowledge (for they are as endless in
number as are the different points of view from which we may
304
perceive and analyse phenomena) are “worth” to-day very largely
only what they have meant and what they may yet mean to human
life, to human thought, to civilization. While there is certainly no
useless truth and no utterly unimportant fact, it is quite possible to
burden and hamper the mind of youth with supposed truths and facts
that have little or no relevancy to any coherent or any real point of
view about human knowledge and human interests either of the past
or the present. It is merely, for example, in the light of the effects that
305
they have had upon the life and thought of humanity that the
great philosophical systems of the past ought (after the necessary
period of preliminary study on the part of the pupil) to be presented
to students in university lectures. A teacher who cannot set them
forth in this spirit is really not a teacher at all—a man who can make
306
his subject live again in the thought of the present.
If the limits of our space and our subject permitted of the
attempt, we might easily continue the study of the pragmatist
element in American scholarship from the point of view of the whole
general economy of a university as a social institution, and from that
of the benefit that has accrued to the modern world from the many
successful attempts at the organization of knowledge from an
international point of view, that have come into being under
307
American initiative.
Lastly it is surely impossible to exaggerate the value to
308
philosophy of the so-called “democratic,” open-minded attitude of
Pragmatism that is seen in its unprejudiced recognition of such
things as the ordinary facts of life, the struggle that constitutes the
309
life of the average man, the fragmentary and partial character of
most of our knowledge, and so on. All this contrasts in the most
favourable way with the scholastic and the Procrustean attitude to
facts that has so long characterized philosophical rationalism from
Leibniz and Wolff to the Kantians and to the Neo-Kantians and the
Neo-Hegelians of our own time. Thanks partly to this direct and
democratic attitude of mind on the part of the pragmatists and
humanists, and thanks too to the entire psychological and
sociological movement of modern times, the points of view of the
different leading thinkers of different countries are beginning to
receive their fitting recognition in the general economy of human
thought to be compared with each other, and with still other possible
points of view.
No one, it seems to me, can read the books of James without
feeling that philosophy can again, as the universal science indeed,
“begin anywhere” in a far less restricted sense than that in which
Hegel interpreted this ingenious saying of his in respect of the
310
freedom of human thinking.
As for the inevitable drawbacks and limitations of the very
Americanism which we have been endeavouring to discover in
Pragmatism, it cannot, to begin with, be entirely without an element
of risk to philosophy, and to the real welfare of a country, that the
highest kind of insight should be brought too ruthlessly into
competition with the various specialized studies, and the various
311
utilitarian pursuits of modern times, and with popular tendencies
generally. The public, for many reasons, should not be too readily
encouraged to think of philosophy as merely “a” study like other
studies and pursuits, to be baited with the idea of its utility and its
profitable consequences. Philosophy, on the contrary, is the
universal study that gives to all other studies and pursuits their
relative place and value. If left too much to be a mere matter of
choice on the part of the young and the unthinking, it will soon find
itself in the neglected position of the wisdom that utters her voice at
the street corners. It must be secured an integral, and even a
necessary place in the world of instruction—a condition that is still
312
the case, it is to be remembered, in Catholic as distinguished
from many so-called “liberal” and “Protestant” seats of learning.
It is possible indeed, as we have already suggested, that the
recognition of an aristocratic or a Catholic element in learning would,
in some respects, be of more true use in the schools of America than
a mere pragmatist philosophy of life and education. And it is
therefore not to be wondered at that Americans themselves should
already have expressed something of a distrust for a philosophy and
an educational policy that are too akin to the practical commercialism
313
of the hour.
Then again, despite the large element of truth that there is in the
idea of philosophy “discovering” (rather than itself “being”) the true
“dynamic” or “motive-awakening” view of the system of things in
which we live, philosophy itself was never intended to bear the entire
weight and strain that are put upon it by the pragmatists. In their
enthusiasm they would make out of it, as we have seen, a religion
(and a new one at that!) and a social philosophy, as well as the
theory of knowledge and the “approach” to reality that we are
accustomed to look for in a system of philosophy.
It is only in periods of transition and reconstruction, like the
present age, when men have become acutely sensible of the
limitations of traditional views of things, that they are inclined in their
disappointment to look to scientific and professional thinkers for
creeds that shall take the place of what they seem for the moment to
be losing. It is in such times chiefly that philosophy flourishes, and
that it is apt to acquire an undue importance by being called upon to
do things that of itself it cannot do. Among the latter impossibilities is
to be placed, for example, the idea of its being able to offer (almost
314
in any sense) a substitute for the direct experience of the
common life, or for the realities of our affections and our emotions, or
for the ideals engendered by the common life.
Owing partly to the limitations of the Intellectualism that has
hitherto characterized so much of the culture and the educational
policy of the last century there are still everywhere scores of people
under the illusion that the truth of life will be revealed to them in the
theory of some book, in the new views or the new gospel of some
emancipated and original thinker. In this vain hope of theirs they are
obviously forgetful of even the pragmatist truth that all theories are
but a kind of transformation, or abstract expression, of the
experiences of real life and of real living. And part of the trouble with
the pragmatists is that they themselves have unwittingly ministered
to this mistaken attitude of mind by creating the impression that their
theory of taking the kingdom of Heaven by storm, by the violence of
their postulations and of their plea for a “working view” of things, is
indeed the new gospel of which men have long been in search. The
race, however, is not always to the swift and the eager, nor the
kingdom to those who are loudest in their cryings of “Lord, Lord.”
And as a friend of mine aptly applied it as against all practicalism
315
and Pragmatism, “there remaineth a rest to the people of God.”
The ordinary man, it should be borne in mind, does not in a certain
sense really need philosophy. Its audience is with the few, and it is to
do it but scant service to think of making it attractive to the many by
the obliteration of most of its distinctive characteristics and
difficulties, and by the failure to point out its inherent limitations. It is
not by any means, as we have been indicating, a substitute either for
life, or for positive religion. Nor can it ever have much of a message,
even for the few, if they imagine themselves, on account of their
wisdom, to be elevated above the needs of the ordinary discipline of
life.
Then again, there is surely an element of considerable danger in
the American-like depreciation of doctrine and theory which we have
noticed in two or three different connexions on the part of
Pragmatism. In the busy, necessitous life of the United States this
316
depreciation is sometimes said to be visible in the great sacrifice
317
of life and energy that is continually taking place there owing to
an unduly literal acceptance on the part of every one of the idea that
each individual has a sort of divine right to seek and to interpret his
experience for himself. In Pragmatism it might be said to be
illustrated in the comparative weakness in the essentials of logic and
ethics to which we have already referred, in the matter of a sound
theory of first principles. And also in its failure to take any really
318
critical recognition of the question of its theoretical and practical
affiliations to tendencies new and old, many or most of which have
long ago been estimated at their true worth and value. Then there is
319
its comparatively superficial interpretation of what is known in the
thought of the day as “Darwinism” and “Evolutionism” and the
endless belief of the unthinking in “progress,” and its failure to see
320
that its very Americanism and its very popularity are things that
are deserving of the most careful study and criticism. What have the
pragmatists left in their hands of their theory, if its mere
“methodology” and its “efficiency-philosophy” and its would-be
enthusiasm were eliminated from it?
Like Americanism in general (which began, of course, as a
revolutionary and a “liberationist” policy), Pragmatism is inclined in
some ways to make too much of peoples’ rights and interests, and
too little of their duties and privileges and of their real needs and
their fundamental, human instincts. It is in the understanding alone of
321
these latter things that true wisdom and true satisfaction are to be
found. And like the American demand for pleasure and for a good
time generally, Pragmatism is in many respects too much a mere
philosophy of “postulations” and “demands,” too much a mere
formulation of the eager and impetuous demands of the
emancipated man and woman of the time—-as forgetful as they of
322
many of the deeper facts of life and of the economy of our human
civilization. In demanding that the “consequences” of all pursuits
(even those of study and philosophy) shall be “satisfying,” and that
philosophy shall satisfy our active nature, it forgets the sense of
disillusionment that comes to all rash and mistaken effort. It certainly
does not follow that a man is going to get certain things from the
world and from philosophy merely because he demands them any
more than does the discovery and the possession of happiness
323
follow from the “right” of the individual to seek it in his own best
way. Nor is it even true that man is called upon to “act” to anything
like the extent contemplated by an unduly enthusiastic Americanism
and an unduly enthusiastic Pragmatism. The writer is glad to be able
to append in this connexion a quotation taken by an American critic
of Pragmatism from Forberg in his criticism of the action-philosophy
of Fichte: “Action, action, is the vocation of man! Strictly speaking,
this principle is false. Man is not called upon to act, but to act justly. If
he cannot act without acting unjustly he had better remain inactive.”
It would not be difficult to match this quotation, or perhaps to
surpass it, with something from Carlyle in respect of the littleness of
man’s claims, not merely for enjoyment, but even for existence; but
we will pass on.
Pragmatism, as we have suggested, certainly falls too readily
into line with the tendency of the age to demand means and
instruments and utilities and working satisfactions, instead of ends
and purposes and values, to demand pleasure and enjoyment
instead of happiness and blessedness. Instead of allowing itself to
do this it should have undertaken a criticism both of the so-called
“wants” of the age, and of the soundness of its own views in respect
of the truth and the happiness that are proper to man as man. There
is a fine epigram of Goethe’s in respect of the limitations of the
revolutionary and the liberationist attitude of those who would seek
to “free” men without first trying to understand them, and to help
them to their true inward development.
Aile Freiheits-Apostel, sie waren mir immer zuwider.
Willkur suchte doch nur jeder am Ende für sich.
Willst du viele befrein, so wag’ es vielen zu dienen.
Wie gefährlich das sey, willst du es wissen?
324
Versuch’s.
Until Pragmatism then makes it clear that it is the free rational
activity, and the higher spiritual nature of man that is to it the norm of
all our thought, and all our activity, and the true test of all
“consequences,” it has not risen to the height of the distinctive
message that it is capable of giving to the thought of the present
time. Unqualified by some of the ideal considerations to which we
have attempted, in its name, and in its interest, to give an
expression, it would not be, for example, a philosophy that could be
looked upon by the great East as the last word of our Western
wisdom or our Western experience. It will be well, however, to say
nothing more in this connexion until we have looked at the
considerations that follow (in our next chapter) upon the lofty, but
impersonal, idealisation of the life and thought of man attempted by
our Anglo-Hegelian Rationalism, and until we have reflected, too,
upon the more feasible form of Idealism attempted in the remarkable
325
philosophy of Bergson, the greatest of all the pragmatists.
CHAPTER VIII
PRAGMATISM AND ANGLO-HEGELIAN
RATIONALISM
The form of Anglo-German Rationalism or Intellectualism which I

shall venture to select for the purposes of consideration from the
point of view of Pragmatism and Humanism is the first volume of the
recent Gifford Lectures of Dr. Bernard Bosanquet, who has long
been regarded by the philosophical public of Great Britain as one of
the most characteristic members of a certain section of our Neo-
Hegelian school. I shall first give the barest outline of the argument
and contentions of “The Principle of Individuality and Value,” and
then venture upon some paragraphs of what shall seem to me to be
relevant criticism.
Dr. Bosanquet’s initial position is a conception of philosophy, and
its task which is for him and his book final and all-determining. To
him Philosophy is (as it is to some extent to Hegel) “logic” or “the
spirit of totality.” It is “essentially of the concrete and the whole,” as
Science is of the “abstract and the part.” Although the best thing in
life is not necessarily “philosophy,” philosophy in this sense of “logic”
is the clue to “reality and value and freedom,” the key to everything,
in short, that we can, or that we should, or that we actually do desire
and need. It [philosophy] is “a rendering in coherent thought of what
lies at the heart of actual life and love.” His next step is to indicate
“the sort of things,” or the sort of “experiences,” or the sort of “facts”
that philosophy needs as its material, if it would accomplish its task
as “universal logic.” This he does (1) negatively, by the rejection of
any form of “immediateness,” or “simple apprehension,” such as the
“solid fact,” the “sense of being,” or the “unshareable self” of which
we sometimes seem to hear, or such as the “naïve ideas” of
326 327
“compensating justice,” “ethics which treats the individual as
328
isolated” and “teleology” as “guidance by finite minds,” as the
data (or as part of the data) of philosophy; and (2) positively, by
declaring that his subject-matter throughout will be “the principle of
‘individuality,’ of ‘self-completeness,’ as the clue to reality.” This
“individuality” or “self-completeness” is then set forth in a quasi-
Platonic manner as the “universal,” the real “universal” being (he
insists) the “concrete universal,” the “whole,” that is to say, “the
logical system of connected members,” that is to him the “ideal of all
thought.” We must think of this “individuality,” therefore, either as “a
living world, complete and acting out of itself, a positive, self-
329
moulding cosmos,” or “as a definite striving of the universe”[!]
The next question (so far as our partial purposes are concerned)
that Dr. Bosanquet asks is, “What help do we get from the notion of a
‘mind’ which ‘purposes’ or ‘desires’ things in appreciating the work of
factors in the universe, or of the universe as [ex-hypothesi] self-
directing and self-experiencing whole?” The answer is spread over
several chapters, and is practically this, that although there is
undoubtedly a “teleology” in the universe (in the shape of the
“conjunctions and results of the co-operation of men,” or of “the
harmony of geological and biological evolution”), and although
“minds such as ours play a part in the work of direction, we cannot
judge of this work in question in any human manner.” The real test of
teleology or value is “wholeness,” “completeness,” “individuality” [the
topic of the book], and it is made quite clear that it is the “Absolute”
who is “real” and “individual” and not we. We are, indeed, in our lives
330
“carried to the Absolute without a break,” and our nature “is only
331
in process of being communicated to us.” “We should not think of
ourselves after the pattern of separate things or personalities in the
legal sense, nor even as selves in the sense of isolation and
exclusion of others.” “Individuality” being this “logical self-
completeness,” there can be only one “Individual,” and this one
Individual is the one criterion of “value,” or “reality,” or “existence,”
“importance” and “reality [!]” being sides of the one “characteristic”
[i.e. “thinkableness” as a whole]. Dr. Bosanquet confesses in his
seventh chapter that this idea of his of “individuality,” or “reality,” is
essentially the Greek idea that it is only the “whole nature” of things
that gives them their reality or value.
We are then assured, towards the close of this remarkable book,
that “freedom” (the one thing that we mortals value as the greatest of
all “goods”) is “the inherent effort of mind considered as a [!] world,
and that the ”Absolute“ [the ”universal” of logic, Plato’s “Idea”] is the
“high-water mark of our effort,” and that each “self” is “more like a
rising and a falling tide than an isolated pillar with a fixed
circumference.” The great fact of the book, the fact upon which its
accomplished author rests when he talks in his Preface of his belief,
“that in the main the work [of philosophy] has been done,” is the daily
“transmutation of experience according to the level of the mind’s
energy and self-completeness,” the continued and the continuous
“self-interpretation [of ‘experience’] through the fundamental principle
of individuality.”
Now it is quite obvious that according to many of the
considerations that have been put forward as true in the foregoing
chapters, this philosophy of Dr. Bosanquet’s which treats the
“concept,” or the “universal” as an end in itself (as the one answer to
all possible demands for a “teleology”) and as an “individual,” “a
perfected and self-perfecting [!] individual,” can be regarded as but
another instance of the abstract Rationalism against which
Pragmatism and Humanism have entered their protests. It is untrue,
therefore, to the real facts of knowledge and the real facts of human
nature. It will be sufficient to state that the considerations of which
we are thinking are (in the main) the positions that have been taken
in respect of such things as: (1) the claim that a true metaphysic
must serve not merely as an intellectual “system” but as a “dynamic,”
and as a “motive” for action and achievement; (2) the fact of the
“instrumental” character of thought and of ideas, and of all systems
(of science or of philosophy or of politics) that fail to include as part
of their data the various ideals of mankind; (3) the idea that all truth
and all thought imply a belief in the existence of objects and persons
independent of the mere mental states or activities of the thinking
individual, and that belief rather than knowledge is, and always has
been, man’s fundamental and working estimate of reality; (4) the fact
that our human actions and re-actions upon reality are a part of what
we mean by “reality,” and that these actions and re-actions of ours
are real and not imaginary; (5) the attitude in general of Pragmatism
to Rationalism; (6) the various concessions that have been made by
representative rationalists to the pragmatist movement.
Dr. Bosanquet’s theory of reality has already impressed some of
his most competent critics as utterly inadequate as a motive or an
incentive to the efforts and endeavours of men as we know them in
history and in actual life, and we shall immediately return to this
topic. And although there are many signs in his Lectures that he is
himself quite aware of the probability of such an impression, his book
proceeds upon the even tenor of its way, following wherever his
argument may lead him, irrespective entirely of the truth contained in
the facts and the positions we have just recounted and reaffirmed. It
lends itself, therefore, only too naturally to our present use of it as a
highly instructive presentation of many, or most, of the tendencies of
Rationalism and Intellectualism, against which Pragmatism and
Humanism would fain protest. At the same time there is in it, as we
332
hope to show, a fundamental element of truth and of fact without
which there could be no Pragmatism and no Humanism, and indeed
no philosophy at all.
333
A broad, pervading inconsistency in “Individuality and Value”
which militates somewhat seriously against the idea of its being
regarded as a tenable philosophy, is the obvious one between the
position (1) that true reality is necessarily individual, and the position
(2) that reality is to be found in the “universal” (or the “concept”) of
334
logic. It would, however, perhaps be unfair to expect Dr.
Bosanquet to effect a harmony between these two positions that
Aristotle (who held them both) was himself very largely unable to do.
There is, in other words, a standing and a lasting contradiction
between any and all philosophy which holds that it is reason [or
logic] alone that attains to truth and reality, and the apparently
natural and inevitable tendency of the human mind [it is represented
in Dr. Bosanquet’s own procedure] to seek after “reality” in the
“individual” thing, or person, or being, and in the perfecting of
“individuality” in God (or in a kingdom of perfected individuals).
The positive errors, however, which we would venture to refer to
as even more fatal to Dr. Bosanquet’s book than any of its incidental
inconsistencies are those connected with the following pieces of
procedure on his part: (1) his manifest tendency to treat the
“universal” as if it were an entity on its own account with a sort of
335
development and “value” and “culmination” of its own; (2) his
tendency to talk and think as if a “characteristic” or a “predicate” (i.e.
the “characteristic” or “quality” that some experiencing being or some
thinker attributes to reality) could be treated as anything at all apart
from the action and the reaction of this “experient” (or “thinker”)
336
conceived as an agent; (3) the tendency to talk of “minds” rather
than persons, as “purposing” and “desiring” things; (4) his tendency
to talk as if “teleology” were “wholeness”; (5) his tendency to regard
(somewhat in the manner of Spinoza) “selves” and “persons” as like
337
“rising and falling tides,” and of the self as a “world of content”
engaged in certain “transformations”; and (6) his tendency to think
and speak as if demonstration [“mediation” is perhaps his favourite
way of thinking of the logical process] were an end in itself, as if we
lived to think, instead of thinking to live.
In opposition to all this it may be affirmed firstly that every
“conception” of the human mind is but the more or less clear
consciousness of a disposition to activity, and is representative, not
so much of the “features” of objects which might appear to be their
“characteristics” from a purely theoretical point of view, as of the
different ways in which objects have seemed to men to subserve the
needs of their souls and bodies. The study of the development of the
“concept” in connexion with the facts of memory and with the slow
evolution of language, and with the “socialized percepts” of daily life
will all tend to confirm this position. The phenomena of religion, for
example, and all the main concepts of all the religions are to be
studied not merely as intellectual phenomena, as solutions of some
of the many difficulties of modern Agnosticism, or of modern
Rationalism, or of modern Criticism, but as an expressive of the
modes of behaviour of human beings (with all their needs and all
their ideals) towards the universe in which they find themselves, and
towards the various beings, seen and unseen, which this universe
symbolises to them. These phenomena and these conceptions are
unintelligible, in short, apart from the various activities and cults and
social practices and social experiences and what not, with which
they have dealt from first to last.
Then it is literally impossible to separate in the manner of Dr.
Bosanquet the “predicate” of thought from the active relations
sustained by things towards each other, or towards the human
beings who seek to interpret these active relations for any or for all
“purposes,” Dr. Bosanquet’s idea, however, of the relation of “mind”
to “matter,” to use these symbols for the nonce (for they are but
338
such), is in the main purely “representational” or
339
intellectualistic. To him “mind” seems to reflect either a “bodily
340
content” or some other kind of “content” that seems to exist for a
“spectator” of the world, or for the “Absolute,” rather than for the man
himself as an agent, who of course uses his memories of himself, or
his “ideal” of himself, for renewed effort and activity. One of the most
important consequences of this unduly intellectualistic view of mind
is that Dr. Bosanquet seems (both theoretically and practically)
341
unable to see the place of “mind,” as “purpose,” in ordinary life, or
342
of the place of mind in evolution, giving us in his difficult but
important chapter on the “relation of mind and body” a version of
things that approaches only too perilously close to Parallelism or
343
Dualism, or even to Materialism. And along with this
quasi-“representational” or “copy-like” theory of mind there are to be
associated his representational and intellectualistic views of the
344 345 346
“self” and the “universal” and “spirit.”
There are, doubtless, hints in Dr. Bosanquet’s pages of a more
347
“dynamic” view of mind or of a deeper view than this merely
“representational” view, but they are not developed or worked into
the main portion of his argument, which they would doubtless very
largely transform. This is greatly to be regretted, for we remember
that even Hegel seemed to notice the splitting-up of the real for our
human purposes which takes place in the ordinary judgment. And of
course, as we have noticed, all “purpose” is practical and theoretical
at one and the same time.
Then, thirdly, it is persons, and not “minds,” who desire and
purpose things, “mind” being a concept invented by the spectator of
activity in a person other than himself, which (from the analogy of his
own conscious activity and experience) he believes to be
348
purposive. Dr. Bosanquet’s use, too, of the expression “mind”
invariably leaves out of the range of consideration the phenomena of
desire and volition—intelligible, both of them, only by reference to an
end that is to be understood from within, and not from outside of the
personality, from the point of view of the mere spectator. The
phenomena of desire and volition are just as integral ingredients of
our lives as persons as are our cognitive states.
Fourthly, it is doubtful whether the treatment of teleology as
“wholeness” (or its sublimation in “Individuality and Value” into
“wholeness”) is much of an explanation of this difficult topic, or
indeed whether it is any explanation at all. Dr. Bosanquet, in fact,
confesses that teleology is a conception which “loses its distinctive
meaning as we deepen its philosophical interpretation, and that it
has very little meaning when applied to the universe as a whole” [a
thing that is apparent to any Kantian student]. “It is impossible
seriously,” he says, “to treat a mind which is the universe[!] as a
workman of limited resources, aiming at some things and obliged to
accept others as means to these.” And it is equally impossible, he
holds, to apply “to the universe” the distinction of “what is purpose for
its own sake and what is not so.” In fact, Dr. Bosanquet’s treatment
of teleology is thus mainly negative, as including not only this
349
rejection of the notion in reference to the “universe as a whole,”
but its rejection, too, in reference to the purposes of our human
350
life; although he admits (as of course he must) that the
conception of end or purpose is drawn from some of the features
(“the simplest features,” he says) of our “finite life,” or “finite
consciousness.” If the notion were “to be retained at all,” he says, “it
could only be a name for some principle which would help to tell us
what has value quite independent of being or not being, the purpose
351
of some mind.” Now, of course, according to the Pragmatism and
Humanism that we have been considering in this book, no intelligent
person could take any conceivable interest in such a useless fancy
as a teleology of this kind. Thus teleology is really blotted out
altogether of existence in this volume, and with its disappearance
there must go also the notion of any value that might be intelligibly
associated with the idea of the attainment of purposes or ends by the
human beings with whom we are acquainted in our ordinary daily
life.
352
We shall below refer to the fact that this rejection of teleology
and value is one that must be regarded as fatal to ethics or to
Absolutism in the realm of ethics. It requires, too, to be added here
that even the most unprejudiced reading of Dr. Bosanquet’s work
must create in the mind of the reader the conviction that its author is
altogether unfair to the views of those who believe in the existence of
353
definite manifestations of purpose in human life. He talks as if
those who uphold this idea or this fact are committed either to the
absurd notion that man is “the end of the universe,” or to the equally
absurd notion that “art, thought, society, history, in which mind begins
to transcend its finiteness should be ascribed to the directive abilities
of units in a plurality, precisely apart from the world content and the
underlying solidarity of spirits, the medium through which all great
things are done.”
With a view of bringing our discussion of these striking Gifford
Lectures within the scope of the general subject of this book the
following might be regarded as their leading, fundamental
characteristics to which the most serious kind of exception might well
354
be taken: (1) its “abstractionism” and its general injustice to fact
355
due to its initial and persistent “conviction” [strange to say, this is
the very word used by Bosanquet] that the real movement in things
is a “logical” movement; (2) its fallacious conception of the task of
philosophy as mainly the obligation to think the world “without
contradiction”; (3) its obvious tendency in the direction of the
356
“subjective idealism” that has been the bane of so much modern
357
philosophy and that is discarded altogether by Pragmatism and
Humanism; (4) its retention of many of the characteristic
358
polemical faults of Neo-Hegelianism and its manifestation of a
359
similar spirit of polemical unfairness on the part of their
accomplished author; (5) its implication in several really hopeless
contradictions in addition to the broad contradiction already referred
to; (6) its failure [a common Neo-Hegelian failing] to do justice to the
spirit and (in certain important regards) the letter of Kant; (7) its
essential non-moralism or its apparently anti-ethical character.
As for the first of these charges, the “abstractionism” of
“Individuality and Value,” coming as it does on the top of the general
perversity of the book, is really a very disastrous thing for philosophy.
360
While we may pardon an enthusiastic literary Frenchman for
saying that, “The fact is, you see, that a fine book is the end for
which the world was made,” there is hardly any excuse for a
philosopher like Dr. Bosanquet coming before the world with the
appearance of believing that the richly differentiated universe that we
know only in part, exists for the benefit of the science that he
represents, for the dialectic of the metaphysician, to enable the
“universal” to “become more differentiated” and “more
361
individualized,” to become “more representative” of the “whole.”
We might compare, says Dr. Bosanquet, in a striking and an
362
enthralling passage, “the Absolute to Dante’s mind as uttered in
the Divine Comedy ... as including in a single, whole poetic
experience a world of space and persons, ... things that, to any
ordinary mind, fall apart.” Now even apart from the highly interesting
question of the manifestly great and far-reaching influence of Dante
over Dr. Bosanquet, and apart, too, from the notable modesty of Dr.
Bosanquet’s confession as to the “imperfect” character of the simile
just reproduced, no one to-day can think of attaching any ultimate
importance to “Dante’s mind” without thinking of the extent to which

Full download Artificial Intelligence and Deep Learning in Pathology 1st Edition Stanley Cohen Md (Editor) file pdf all chapter on 2024

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Full download Artificial Intelligence and Deep Learning in Pathology 1st Edition Stanley Cohen Md (Editor) file pdf all chapter on 2024

Uploaded by

Copyright:

Available Formats

Artificial Intelligence and Deep

Learning in Pathology 1st Edition

Risk Modeling: Practical Applications of Artificial

Pro Deep Learning with TensorFlow 2.0: A Mathematical

Artificial Intelligence and Machine Learning for EDGE

Artificial Intelligence and Machine Learning in the

Artificial Intelligence and International Relations

Predicting Heart Failure : Invasive, Non-Invasive,

Visualization Techniques for Climate Change with

Artificial Intelligence for Dummies 2nd Edition John

Stanley Cohen, M.D

Copyright © 2021 Elsevier Inc. All rights reserved.

Library of Congress Cataloging-in-Publication Data

British Library Cataloguing-in-Publication Data

For information on all Elsevier publications visit our website at

Publisher: Dolores Meloni

It is now well-recognized that artificial intelligence (AI) represents an inflection

disciplines. The purpose of this monograph is to provide a detailed understanding of

The evolution of machine

Artificial Intelligence and Deep Learning in Pathology. https://doi.org/10.1016/B978-0-323-67538-3.00001-4 1

Rules-based versus machine learning: a deeper look

Varieties of machine learning

All of those latter approaches, in addition to neural networks, are collectively

General aspects of machine learning

Deep learning and neural networks

The role of AI in pathology

In many ways, the experiences of radiologists are helping to guide us in the

efficient than current fully software implementations such as both conventional

The form of Anglo-German Rationalism or Intellectualism which I

You might also like