Professional Documents
Culture Documents
Cat Grading Review
Cat Grading Review
CURRENT
OPINION Cataract grading systems: a review of past
and present
Helena E. Gali, Ruti Sella, and Natalie A. Afshari
Purpose of review
To provide a comprehensive summary of past cataract grading systems, how they have shaped
current grading systems, and the developing technologies that are being used to assess and grade
cataracts.
Recent findings
This summary of cataract grading systems examines the development and limitations that existed in past
grading systems and how they have shaped the grading systems of present time. The Lens Opacities
Downloaded from http://journals.lww.com/co-ophthalmology by BhDMf5ePHKbH4TTImqenVIiuKVF7qTxsO7egTcR85zhfbHf7wqjCQCHIL2A3JkMwaD1TuyJxyDU= on 01/10/2019
Classification System III (LOCS III) system is currently used both clinically and for research purposes. Recent
advancements in imaging technologies have allowed researchers to create automatic systems that can
locate lens landmarks and provide cataract grading scores that correlate well with LOCS III clinical grades.
Utilizing existing technologies, researchers demonstrate that fundus photography and optical coherence
tomography can be used as cataract grading tools. Lastly, deep learning has proved to be a powerful tool
that can provide objective and reproducible cataract grading scores.
Summary
Cataract grading schemes have provided ophthalmologists with a way to communicate clinical findings
and to compare new developments in diagnostic technologies. As technologies advance, cataract grading
can become more objective and standardized, allowing for improved patient care.
Keywords
automatic systems, cataract grading, deep learning, Lens Opacities Classification System III, optical
coherence tomography
1040-8738 Copyright ß 2018 Wolters Kluwer Health, Inc. All rights reserved. www.co-ophthalmology.com
&&
nuclear opacification in addition to adding color user variation in the slit lamp examination [2 ].
photographs to be used as standards for comparison This comparison marked an important change in
[8]. Visual acuity was again left out of the grading welcoming the idea that photographic assessment
scheme, as there are other factors that contribute to could be superior to the slit lamp exam, perhaps
changes in vision that could be unrelated to lens creating a gateway for digital analysis to provide a
abnormality. LOCS II demonstrated good intra- more objective measure of cataract than the subjec-
observer and inter-observer agreements at the slit tive clinician assessment. At present, the LOCS III
lamp and in photographic readings [8]. The slit lamp system is still used in varying degrees in clinical
exam proved to be slightly more sensitive at detect- practice, however, its clinical impact on decision-
ing opacification than photographs, likely owing to making and timing of surgery is questionable.
limitations in photographic technologies at the
time [8].
The LOCS III was created in 1993 to improve CORRELATION BETWEEN GRADE AND
upon limitations in LOCS II. LOCS III expanded the VISUAL SYMPTOMS FOR THE PATIENT
scales used in LOCS II to better capture the early Before discussing cataract grading systems further, an
stages of cataract formation in the grading scheme important question to ask is how it relates to the
&&
[2 ]. It examines nuclear opalescence (NO) and symptoms experienced by the patient. If cataract
nuclear color (NC) on a scale from 1 to 6, cortical surgery is to be performed regardless of cataract
cataracts (C) on a scale from 1 to 5, and posterior severity and instead based on patient experience of
subcapsular cataracts on a scale from 1 to 5 (Fig. 1) their cataract, what benefit does a comprehensive
&&
[2 ]. Although a finer grading scheme lowers con- grading scheme provide? From an epidemiological
cordance among measurements, the resulting standpoint, it is important to characterize cataract
increase in sensitivity was chosen as a worthy reason type and severity to better understand population
for the change. Of note, the original 1993 article health and treatment need, but how do cataract
highlighting improvements of LOCS III over LOCS II grades perform on a patient-to-patient basis? In a
compared photographic evaluation of cataracts as 2012 study examining LOCS III grading and visual
opposed to in-vivo evaluation based on the ability to functioning, it was noted that vision-specific func-
standardize photographic protocols and to reduce tioning decreased in a statistically significant way for
FIGURE 1. The Lens Opacities Classification System III Standards introduced in 1993. Data from [2 ]. &&
1040-8738 Copyright ß 2018 Wolters Kluwer Health, Inc. All rights reserved. www.co-ophthalmology.com 15
different cataract types at different grades using III system was introduced, Hall et al. set out to
&
the LOCS III [9 ]. Overall, the trend makes sense in provide a new way to evaluate nuclear cataracts
that as grade increased, vision-specific functioning using a laser slit lamp in an attempt to standardize
decreased. The nuance, however, lies in the fact that an objective way to visualize cataract complexities.
each cataract type produced symptoms at different The team created a device using a laser light slit,
grades. Nuclear cataracts affected functioning at viewing arm with a beam splitter, and a charge
grades 4 for opalescence and 5 for color, cortical coupled device camera to illuminate the anterior
cataracts affected functioning at grade 3, and poste- segment with laser light [11]. The images obtained
rior subcapsular cataracts affected functioning at all with the laser were analyzed with a computer pro-
&
grades [9 ]. Although it may make sense to a clinician gram that calculated mean pixel intensity of the area
that a posterior subcapsular cataract produces visual of interest in the nucleus as compared with a darker
symptoms in a more significant way than does a background portion of the anterior segment in the
nuclear cataract, from a patient perspective it may same image to measure the amount of light back-
be confusing to find that a cataract grade of P1 is scattered from the nucleus [11]. This method proved
much more visually bothersome than a cataract grade to correlate well with the LOCS III grading scheme,
of NC3 if the patient is unfamiliar with the nuances of demonstrating a linear relationship between
cataract disease. The study raises an important con- nuclear opalescence LOCS III scores and the pixel
sideration of whether or not vision-specific function- intensities imaged by the laser slit lamp and ana-
ing should be incorporated back into the assessment lyzed with computer software [11]. This highlighted
of cataracts if the grading scheme is not only meant to the use of a computer program to assist with
incorporate physician findings but also patient per- cataract grading.
spective and understanding. Following the work of Hall, Babizhayev et al.
[12] set out to use computer generated analysis of
lens imaging to measure cataract severity in combi-
CORRELATION BETWEEN GRADE, nation with a glare disability test. An automated
SURGICAL APPROACH, AND OUTCOME computer program was used to measure the inten-
Not only is it important to consider the patient’s sity of individual image pixels, and using serial 2D
experience of their cataracts and how it relates to images, 3D topography images of the lens were
cataract grade, it is also important to consider how created to provide a better understanding of the
the surgical approach and outcome are determined characteristics of the lens [12]. The computer’s
by cataract grading. Will surgical approach be dif- analysis of the lens was compared against a clinical
ferent if a cataract is given a grade of NO3 and NC3 slit lamp examination using the same grading
versus NO4 and NC4? A study by Davidson and scheme laid out by Taylor and West. This analysis
Chylack investigating the use of LOCS III and pha- technique demonstrated correlation between the
coemulsification performance in 2364 cases found existing LOCS III scoring system and a new way
that nuclear cataract phacoemulsification time cor- to assess cataracts [12].
related well with LOCS III but that cortical and Additional teams of researchers further refined
posterior subcapsular phacoemulsification had little automatic systems for classifying cataracts using slit
&&
relation to LOCS III classes [10 ]. Of note, an expo- lamp photography. Fan et al. [13] created a method
nential increase in phacoemulsification energy was to grade nuclear sclerotic cataracts based on land-
used intraoperatively as nuclear grades increased, marks in the visual axis from slit lamp photographs.
proving LOCS III to be a useful tool in creating an Another group from Johns Hopkins also used photo-
&&
operative plan for nuclear cataract procedures [10 ]. graphs from the visual axis and analyzed a variety of
Cataract scores can play a useful role in determining features of the lens [14]. In these methods, a neural
case complexity during cataract surgery and post- network was employed to analyze the images and
operative outcomes. This preoperative knowledge produce a grade of nuclear cataract severity based on
can better allow clinicians to allocate certain cases the lens landmarks noted in the visual axis [14]. At
to surgeons in training and to provide realistic risks the time, these models focused only on the visual
and potential outcomes to patients. axis, whereas a clinical exam evaluated the entirety
of the lens.
Li et al. [15] expanded on previous groups’ work
NEW WAYS TO ASSESS AND GRADE to create an automatic diagnosis system for nuclear
CATARACTS cataracts by training the system to locate the entire
Since the development of the LOCS III system, there lens and to extract features such as intensity, color,
have been a variety of ways presented to better and entropy inside the lens and nucleus. They based
evaluate and grade cataracts. Shortly after the LOCS the lens structure detection ability of their program
on their prior work with the Active Shape Model but they may provide alternatives to cataract grad-
method that they described in previous articles, and ing when resources or time are limited. Using imag-
they built upon it to better identify the lens nucleus ing modalities in addition to slit lamp photographs,
&&
in addition to the lens itself [16,17]. The lens and Xiong et al. [20 ] suggest that retinal images can be
feature detection system was tested on over 5800 used to screen for cataracts. They employed the scale
images with a 96.8% lens location detection rate and used by Wang et al. [21] to grade blurriness in retinal
a 95% lens structure detection rate [15]. When images. Previous work on the topic was performed
compared against clinical diagnoses of the same with Fourier analysis by Abdul-Rahman et al. [22].
images, the automatic diagnosis system only varied Their automated method of quantifying optical
by greater than 1 clinical grade in only 3.37% of degradation in retinal images proved to be useful
images reviewed, proving its agreement with clinical in detecting the presence of cataract and correlated
grading schemes and clinical diagnostic methods well with LOCS III [22]. However, automatic grading
[15]. Xu et al. [18] set out to create an automatic of the cataract was not performed. Additionally,
grading system that refined the work of Li by build- vitreous opacity could not be evaluated separately
ing on the success of the lens structure detection and from lens opacities, potentially influencing grading
identifying the most effective elements from the [22]. With these limitations in mind, Xiong et al.
image. This bag-of-features (BOF) method allows developed a successful method to detect vitreous
for location-independent representation of features opacity and separate it from the retinal structure
that can be assessed for qualities such as intensity detection so as to improve cataract grading accu-
and scale. Combining the BOF method with regres- racy. This system graded cataracts with an 81.1%
sion model training, the system achieved a 69.0% accuracy and kappa value of 0.7435 when compared
&&
exact agreement ratio when compared with clinical with clinical grading [20 ]. As technologies in tele-
grading and a 98.9% decimal grading error less than medicine and artificial intelligence advance, using
1.0 [18]. These results demonstrated that clinical retinal images to grade cataract may prove an effi-
diagnosis could be facilitated with the use of auto- cient and useful tool to utilize fundus photos that
matic diagnosis systems. may already be taken for diabetic retinopathy or age-
Improving upon these automatic diagnosis sys- related macular degeneration monitoring.
tems, Srivastava et al. [19] developed what they call In addition to fundus photography being used
the Automatic Cataract Screening from Image Anal- as a novel method to grade cataracts, optical coher-
ysis-Nuclear Cataract, Version 0.10, ACASIA- ence tomography (OCT) has also been suggested.
NC_v0.10. The system builds upon the work done Anterior segment OCT has been used to assess a
by Li et al. in that it uses image gradients on edges variety of features such as corneal thickness, but
produced by landmarks in the lens. At higher grades Wong et al. thought to compare the anterior seg-
of nuclear cataract, these landmarks tend to be less ment OCT nucleus density measurement with LOCS
visible as the landmarks become less distinct. The III grading of nuclear opalescence and nuclear color.
ACASIA-NC_v0.10 instead focuses on the gradient They found that there was a significant correlation
between areas of the lens as a means to measure the between the two measurements and that nuclear
difference in grades of cataract and to overcome the opalescence scores had a slightly higher association
loss of visible landmarks in severe cataracts [19]. with anterior segment OCT than did nuclear color
Although the ACASIA-NC_v0.10 did not prove to [23]. Their use of anterior segment OCT to grade lens
be useful in grading very low or very high grade density proved to be an objective, reliable, and fast
nuclear cataracts, it improved upon prior methods assessment using a frequently used clinical tool that
that used color and intensity and demonstrated that requires less training than mastering the LOCS III.
gradient information is useful in automatic cataract
grading programs [19].
THE ROLE OF DEEP LEARNING IN
CATARACT GRADING
BEYOND SLIT LAMP PHOTOGRAPHY: A discussion of cataract grading would be incom-
NEW TECHNIQUES IN CATARACT plete without touching on recent advancements in
ASSESSMENT AND GRADING the use of deep learning with machine learning and
Although slit lamp examination and slit lamp pho- artificial intelligence to assess lens opacity.
tographs have been the mainstay of cataract grading Although the existing automatic methods to grade
systems, different ways to assess for cataracts have cataracts discussed previously rely on predefined
been proposed. These methods may vary in the landmarks and features to recognize structures
exactness of which they can assess cataracts as com- and grade cataracts, systems can be trained to learn
pared with standard slit lamp photography analysis, grading features, filter them, and feed them into a
1040-8738 Copyright ß 2018 Wolters Kluwer Health, Inc. All rights reserved. www.co-ophthalmology.com 17