State of Art

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Image‌‌classification‌‌

 
approach‌‌for‌‌malaria‌‌ 
detection‌  ‌
 ‌

State‌‌of‌‌art‌  ‌
 ‌

Malaria‌‌disease‌‌is‌‌always‌‌nowadays‌‌in‌‌actuality‌‌and‌‌costs‌‌1millon‌‌of‌‌life‌‌according‌‌to‌‌the‌‌World‌‌
 
Health‌‌Organization‌‌(WHO).‌‌Most‌‌of‌‌the‌‌study‌‌of‌‌diagnostics‌‌is‌‌oriented‌‌to‌‌automatics‌‌detection‌ 
of‌‌plasmodium‌‌and‌‌different‌‌species.‌  ‌

In‌‌our‌‌research,‌‌several‌‌researchers‌‌worked‌‌on‌‌computer-assisted‌‌disease‌‌detection.‌  ‌

We‌‌have‌‌to‌‌make‌‌a‌‌review‌‌of‌‌different‌‌methods‌‌of‌‌malaria‌‌disease‌‌automatic‌‌detection.‌‌We‌‌are‌‌
 
going‌‌to‌‌process‌‌like‌‌in‌‌paper‌‌[Sumi‌‌et‌‌al‌‌2020]‌‌at‌‌suite‌‌of‌‌study.‌  ‌

 ‌
Tableau‌1
‌ :‌‌‌The‌‌list‌‌of‌‌selected‌‌papers‌‌of‌‌automatic‌‌detection‌‌of‌‌Plasmodium‌‌parasites‌  ‌

ID‌  ‌ AUTHOR‌  ‌ Topics‌  ‌ Method/Contrib‌ Datasets‌  ‌ Performance‌  ‌


ution‌  ‌

1‌  ‌ Budi‌‌Sunarko‌‌et‌‌al‌‌
  Segmentation‌  ‌ Otsu‌‌
  Kaggle‌‌hosting‌‌a‌‌repository‌  ‌ Accuracy‌‌94.60%,‌‌ 
2020‌  ‌ method/SVM‌  ‌ specificity‌‌96.20%‌‌ 
sensitivity‌‌93%‌  ‌
2‌  ‌ Tehreem‌‌Fatima‌‌1‌‌
  classification‌  ‌ bilateral‌‌  publicly‌‌available‌‌at‌  ‌ accuracy‌‌is‌‌0.9180,‌‌ 
•‌‌Muhammad‌‌  filtering/object‌‌  specificity‌‌0.95%‌‌ 
Shahid‌‌Farid‌‌2020‌  ‌ contours/8-connec‌ https://ceb.nlm.nih.gov/repo‌ and‌‌precision‌‌ 
ted‌‌rule‌  ‌ sitories/malaria-datasets/‌   ‌ 0.9466‌  ‌

3‌  ‌ K.‌‌M.‌‌Faizullah‌‌  classification‌  ‌ CNN-SVM‌‌/‌‌


  public‌  ‌ Accuracy‌‌0.9923,‌‌ 
Fuhad‌‌et‌‌al‌‌2020‌  ‌ CNN-KNN‌  ‌ Sensitivity‌‌0.9952,‌‌ 
Specificity‌‌0.9917,‌‌ 
Precision‌‌0.9892‌  ‌

4‌  ‌ Asma‌‌Maqsood‌‌et‌‌
  Data‌‌  customized‌‌CNN‌‌
  NIH‌‌Malaria‌‌Dataset‌  ‌ Specificity‌‌0.9778,‌‌ 
al‌‌2021‌  ‌ Preprocessing/‌‌  model‌  ‌ Sensitivity‌‌0.9633,‌  ‌
Segmentation/‌‌ 
Feature‌‌  Precision‌‌0.9682,‌‌
 
Extraction/‌  ‌ Accuracy‌‌0.9682‌  ‌

classification‌  ‌

5‌  ‌ W‌‌Swastika‌‌*‌‌,‌‌G ‌‌ classification‌  ‌ Image‌‌Gray-world‌‌  The‌‌dataset‌‌contains‌‌a‌‌total‌‌  Specificity‌‌0.945,‌  ‌


M‌‌Kristianti‌‌and‌‌R ‌‌ preprocessing‌‌/ ‌‌ of‌2‌ 7,558‌‌cell‌‌images‌  ‌
B‌‌Widodo‌‌(2021)‌  ‌ Customized‌‌CNN‌  ‌ Sensitivity‌‌0.922,‌  ‌

Accuracy‌‌0.931‌  ‌

6‌  ‌ Comparison‌‌of‌‌  classification‌  ‌ CNN‌‌Inceptionv3‌  ‌ https://www.kaggle.com/iaru‌ Accuracy‌‌96.09%‌  ‌


CNN‌‌and‌‌SVM‌‌in‌  nava/cell-images-for-detecti‌
detection‌  ‌ /‌  ‌ ng-malaria‌  ‌ F-Score‌‌of‌0
‌ .97%‌  ‌

of‌‌activation‌‌in‌‌  SVM‌  ‌ /‌  ‌


malaria‌‌cell‌‌
  Accuracy‌‌86.50‌  ‌
images(2019)‌  ‌
F-Score‌‌of‌‌0.89%‌  ‌
7‌  ‌ Malaria‌‌detection‌‌  classification‌  ‌ customized‌‌CNN‌‌   https://ceb.nlm.nih.gov/repo‌ Accuracy‌‌95.32%,‌  ‌
from‌‌RBC‌‌images‌‌  model‌‌VS‌‌  sitories/malaria-datasets‌  ‌
using‌‌shallow‌‌  VGG-16‌‌and‌‌   Sensitivity‌‌ 
Convolutional‌‌  Resnet-50‌  ‌ 94.30%,‌  ‌
Neural‌‌ 
Specificity‌‌ 
Networks(2020)‌  ‌
96.34%,‌  ‌

Precision‌‌96.26%‌  ‌

8‌  ‌ A‌‌Novel‌‌Stacked‌‌   classification‌  ‌ Image‌‌


  The‌‌dataset‌  ‌ accuracy‌‌99.964%,‌‌  
CNN‌‌for‌‌Malarial‌‌   preprocessing‌‌/ ‌‌ precision‌‌100.0%,‌  ‌
Parasite‌‌Detection‌‌  Stacked‌‌CNN‌  ‌ contains‌‌27,‌‌558‌‌images‌  ‌
in‌‌Thin‌‌Blood‌‌  recall‌‌99.928%,‌‌  
Smear‌‌  and‌‌F1-‌‌score‌‌ 
Images(2020)‌  ‌ 99.964%‌  ‌

9‌  ‌ Saturation‌‌channel‌‌  Segmentation‌  ‌ Pre-processing/‌  ‌ image‌‌datasets‌‌of‌‌25‌‌infected‌‌  Accuracy‌‌99.83%,,‌  ‌


extraction‌‌of‌‌HSV‌‌  red‌‌blood‌‌cells‌  ‌
segmentation/‌  sensitivity‌8
‌ 5.06%,‌ 
color‌‌space‌‌for‌‌ 
segmenting‌  ‌ improvement‌‌of‌‌ 
specificity‌‌99.90‌  ‌
segmentation‌‌ 
Plasmodium‌  results/‌  ‌
parasite(2020)‌  ‌
evaluation‌‌
 
processes‌  ‌

10‌  ‌ Malaria‌‌Detection‌‌
  Segmentation/c‌ Local‌‌Composite‌‌
  The‌‌dataset‌  ‌ 63.25%‌  ‌
Using‌‌Local‌‌  lassification‌  ‌ Pattern/SVM‌  ‌
contains‌‌27,‌‌558‌‌images‌  ‌
Composition‌‌ 
Pattern(2021)‌  ‌

11‌  ‌ Malarial‌‌Parasite‌‌
  classification‌  ‌ Logistic‌‌
  The‌‌dataset‌  ‌ Accuracy‌‌66.3643‌  ‌
Detection‌  ‌ regression/‌  ‌
contains‌‌27,‌‌558‌‌images‌  ‌ Accuracy‌‌67.6161‌  ‌
by‌‌Leveraging‌‌
  Random‌‌forest/‌  ‌
Cognitive‌‌  Accuracy‌‌68.9949‌  ‌
Algorithms:‌  ‌ Decision‌‌tree/‌  ‌
Accuracy‌‌59.0348‌  ‌
A‌‌Comparative‌‌
  Naïve‌‌Bayes/‌ 
Accuracy‌‌96.20‌  ‌
Study(2021)‌  ‌ CNN/‌  ‌
Accuracy‌‌94.82‌  ‌
MobileNet‌  ‌

12‌  ‌ Detection‌‌and‌‌  Classification‌  ‌ Normalized‌  ‌ 160‌‌Images‌‌collected‌‌from‌‌   Accuracy‌‌95%‌  ‌


Classification‌‌of‌‌  the‌‌Center‌‌for‌‌Disease‌  ‌
Malaria‌‌in‌‌Thin‌‌  Cross‌‌Correlation‌‌
  Classifying‌‌100%‌  ‌
Blood‌‌Slide‌  NCC‌‌function‌  ‌
Images(2017)‌  ‌

13‌  ‌ A‌‌Novel‌‌  Classification‌  ‌ customized‌‌CNN‌‌


  The‌‌dataset‌  ‌ Accuracy‌‌95.28%‌  ‌
Implementation‌‌of‌‌  model‌  ‌
Deep-Learning‌‌   contains‌‌27,‌‌558‌‌images‌  ‌ Specificity‌‌0.955‌  ‌
Approach‌‌on‌‌ 
Sensitivity‌‌0.950‌  ‌
Malaria‌‌Parasite‌‌
 
Detection‌‌from‌‌  Precision‌‌0.955‌  ‌
Thin‌‌Blood‌‌Cell‌ 
Images(2021)‌  ‌
14‌  ‌ Malaria‌‌Detection‌‌   Classification‌  ‌ CNN-SVM‌  ‌ The‌‌dataset‌  ‌ accuracy‌‌of‌‌98.8%‌  ‌
on‌‌Giemsa-stained‌‌ 
blood‌‌smears‌‌  contains‌‌27,‌‌558‌‌images‌  ‌ specificity‌‌97.6%‌  ‌
using‌‌Deep‌‌ 
sensitivity‌‌98.3%‌  ‌
Learning‌‌and‌‌  
Feature‌‌ 
Extraction(2020)‌  ‌

15‌  ‌ Analyzing‌‌Malaria‌‌  Classification‌  ‌ CNN(Xception/In‌ 7000‌‌images‌  ‌ accuracy‌‌of‌‌ 


Disease‌‌Using‌‌
  ception-V3/ResN‌ 99.28%‌  ‌
Effective‌‌Deep‌‌
  et-50/NasNetMob‌
Learning‌‌  ile/VGG-16/Alex‌ Precision‌‌99.29%‌  ‌
Approach(2019)‌  ‌ Net)‌  ‌

 ‌

 ‌

In‌‌these‌‌papers,‌‌performance‌‌of‌‌methods‌‌used‌‌to‌‌resolve‌‌problems‌‌is‌‌evaluated‌‌following‌‌four‌‌ 
concepts‌‌such‌‌as‌‌accuracy,‌‌precision,‌‌sensitivity‌‌and‌‌specificity[Amin‌‌Siddiq‌‌Sumi‌‌et‌‌al‌‌2021].‌  ‌

There‌‌were‌‌three‌‌evaluation‌‌methods‌‌employed,‌‌i.e.,‌‌accuracy,‌‌sensitivity,‌‌and‌‌specificity.‌‌The‌‌  
accuracy‌‌was‌‌used‌‌to‌‌calculate‌‌the‌‌chance‌‌of‌‌correctness‌‌of‌‌these‌‌results‌‌to‌‌detect‌‌which‌‌one‌‌was‌‌
 
the‌‌Plasmodium‌  ‌

and‌‌which‌‌one‌‌was‌‌not.‌‌The‌‌sensitivity‌‌was‌‌the‌‌chance‌‌that‌‌this‌‌method‌‌could‌‌detect‌‌the‌‌ 
Plasmodium‌‌well.‌‌The‌‌specificity‌‌was‌‌the‌‌chance‌‌that‌‌this‌‌method‌‌was‌‌able‌‌to‌‌detect‌‌ 
non-Plasmodium‌‌objects.‌‌Other‌‌comparisons‌‌that‌‌we‌‌can‌‌find‌‌are‌‌image‌‌processing‌‌methods‌‌and‌‌ 
deep‌‌learning‌‌methods.‌‌The‌‌origin‌‌and‌‌size‌‌of‌‌the‌‌data‌‌are‌‌also‌‌criteria‌‌to‌‌differentiate‌‌the‌‌ 
algorithms.‌  ‌

 ‌

Review‌ 
 ‌

Based‌‌on‌‌several‌‌latest‌‌research‌‌works,‌‌an‌‌artificial‌‌intelligence‌‌approach‌‌based‌‌on‌‌machine‌‌
 
learning‌‌and‌‌deep‌‌learning‌‌algorithms‌‌such‌‌as‌‌Logistic‌‌regression,‌‌Random‌‌forest,‌‌Decision‌‌tree,‌‌ 
Naïve‌‌Bayes,‌‌CNN‌‌andMobileNet‌‌,‌‌has‌‌been‌‌widely‌‌used‌‌and‌‌successfully‌‌obtained‌‌the‌‌level‌‌of‌‌ 
accuracy,‌‌sensitivity,‌‌specificity,‌‌and‌‌precision‌‌of‌‌more‌‌than‌‌90%‌‌[Raghav‌‌Sharma‌‌et‌‌al‌‌
 
2021][Amin‌‌Siddiq‌‌Sumi‌‌et‌‌al‌‌2021].‌  ‌

The‌‌fundamental‌‌difference‌‌between‌‌machine‌‌learning‌‌and‌‌deep‌‌learning‌‌algorithms‌‌is‌‌the‌‌  
feature‌‌extraction‌‌process.‌‌When‌‌machine‌‌learning‌‌performs‌‌step‌‌feature‌‌selection,‌‌deep‌‌learning‌‌  
does‌‌it‌‌automatically‌‌through‌‌trainable‌‌feature‌‌extraction.‌‌In‌‌addition,‌‌machine‌‌learning‌‌needs‌‌  
smaller‌‌datasets;‌‌the‌‌training‌‌process‌‌is‌‌faster,‌‌and‌‌the‌‌accuracy‌‌is‌‌plateauing‌‌but‌‌deep‌‌learning‌‌
 
requires‌‌a‌‌larger‌‌dataset‌‌and‌‌a‌‌long‌‌training‌‌time.‌‌Therefore,‌‌to‌‌have‌‌best‌‌performance‌‌and‌‌  
overcome‌‌the‌‌weaknesses‌‌of‌‌machine‌‌learning‌‌and‌‌deep‌‌learning‌‌algorithms,‌‌some‌‌researchers‌‌in‌‌  
the‌‌field‌‌of‌‌detection‌‌of‌‌parasite‌‌Plasmodium‌‌are‌‌currently‌‌using‌‌an‌‌image‌‌processing‌‌approach‌‌ 
and‌‌combining‌‌it‌‌with‌‌machine‌‌learning‌‌and‌‌deep‌‌learning‌‌algorithms.‌  ‌
 ‌

You might also like