Table 3.
Summary of existing works related to ML and DL techniques for medical imaging
| S.No. | Article | Year | Technique(s) | Parameter(s) | Identified Disease(s) | Performance Metric(s) |
|---|---|---|---|---|---|---|
| Cancer | ||||||
| 1 | [104] | 2016 |
● Weighting-based TL approach ● Supervised learning ● SVM with a gaussian kernel |
● Maximum mean discrepancy |
● Lung cancer ● Brain disease |
- |
| 2 | [23] | 2018 |
● Radiomics ● Extension of CAD ● CNN ● DL ● TL |
● Tumor signatures ● Features extracted from radiomics |
● Breast cancer | - |
| 3 | [81] | 2019 |
● CNN ● Image segmentation ● Feature selection using information retrieval techniques. ● Regression using RMSE ● Clustering |
● Object detection ● Pattern recognition ● Image classification |
● Cancer | - |
| 4 | [44] | 2020 |
● DL ● Semi-supervised learning |
● Labeled data ● Loss functions ● Data re-weighed |
● Multiple disease (breast lesion detection, cancer detection) | - |
| 5 | [103] | 2020 | ● CAD based on CNN |
● AUC ● Sensitivity ● Specificity ● Multiview features ● Five human diagnostics |
● Breast cancer classification (benign and malignant) |
● Sensitivity: 88.6% ● Specificity: 87.6% ● AUC: 0.9468 |
| 6 | [86] | 2020 |
● Robust DL ● CAD tool ● Big data ● TL ● Interpretable AI |
● Clinical data | ● Lesion detection | - |
| 7 | [39] | 2020 |
● DL ● ML |
● Accuracy ● FMeasure ● AUC ● Precision |
● Breast cancer |
DDSM dataset: ● Accuracy: 97.4% ● AUC: 0.99 Inbreast dataset: ● Accuracy: 95.5% ● AUC: 0.97 BCDR dataset: ● Accuracy: 96.6% ● AUC: 0.96 |
| 8 | [38] | 2021 |
● Imagescope (Aperio Imagescope) ● Normalized median intensity |
● Color appearance matrix ● Annotated image |
● Pathology ● Cancer analysis |
Quality performance: ● QSSIM: 97.59% ● SSIM: 98.22% ● PCC: 98.43% |
| Tumor | ||||||
| 9 | [84] | 2016 |
● ANN ● RF ● SVM ● 10-fold cross-validation |
● Image feature | ● Breast tumor |
Accuracy: ● SVM: 77.7% ● RF: 78.5% |
| 10 | [56] | 2017 |
● DL ● Feature selection algorithm ● Pooling ● 2D-CNN ● Big data |
● Numerical or nominal values |
● Lung tumor ● Brain disease |
- |
| 11 | [82] | 2019 |
● ML ● DL ● MRI |
● Brain tissues | ● Brain tumor | - |
| 12 | [102] | 2020 |
● Pixel intensity ● Filtering ● Side detection ● Segmentation ● FLASH (reduction of red eye from images) |
● Human face | ● Brain tumor | - |
| 13 | [97] | 2020 |
● CAD ● TL ● Fuzzy feature selection ● Correlation feature selection |
● Hand crafted features | ● Breast tumors (benign and malignant) |
Accuracy: ● Benign: 100% ● Malignant: 96% |
| Multiple disease | ||||||
| 14 | [42] | 2014 |
● Fusion algorithms ● Morphological knowledge ● Neural network ● Fuzzy logic ● SVM |
● Principal components feature ● Wavelets |
● Brain ● Breast ● Prostate ● Lungs |
- |
| 15 | [29] | 2017 |
● CAD ● Naïve bayes ● SVM ● Functional trees |
● 13 features from 76 features |
● Heart ● Diabetes ● Liver ● Dengue ● Hepatitis |
Accuracy: ● Heart disease using SVM: 94.6% ● Diabetes using Naïve bayes: 95% ● Liver disease using functional tree: 97.1% ● Hepatitis disease using feed forward neural network: 98% ● Dengue disease using rough set theory: 100% |
| 16 | [33] | 2020 |
● Radiography ● MIL |
● Imaging annotation |
● Lung cancer ● Breast cancer |
● Unsupervised feature learning: 93.56% ● Fully supervised feature learning: 94.52% ● MIL performance of coarse label: 96.30% ● Supervised performance of fine label: 95.40% |
| 17 | [32] | 2020 | ● ML |
● Trained models ● Human expert’s narrows |
● Integrated disease | - |
| 18 | [23] | 2020 |
● Supervised and unsupervised ML algorithms ● DT ● Bootstrap methods |
● Clinical data | ● Multiple diseases | - |
| Skin disease | ||||||
| 19 | [4] | 2019 | ● Tensorflow inception version-3 |
● Sensitivity ● Specificity ● PPV, NPN, MCC ● F1 score |
● Acne ● Atopic dermatitis ● Impetigo ● Psoriasis ● Rosacea |
Acne: Sensitivity: 73.3%, Specificity: 95%, PPV: 78.6%, NPV: 93.4%, MCC: 70.1%, F1 score: 75.9% Atopic dermatitis: Sensitivity: 63.3%, Specificity: 87.5%, PPV: 55.9%, NPV: 90.5%, MCC: 48.6%, F1 score: 59.4% Impetigo: Sensitivity: 63.3%, Specificity: 93.3%, PPV: 70.4%, NPV: 91.1%, MCC: 59%, F1 score: 66.7% Psoriasis: Sensitivity: 66.7%, Specificity: 89.2%, PPV: 60.6%, NPV: 91.5%, MCC: 53.9%, F1 score: 63.5% Rosacea: Sensitivity: 60%, Specificity: 91.7%, PPV: 64.3%, NPV: 90.2%, MCC: 53%, F1 score: 62.1% |
| Diabetes | ||||||
| 20 | [76] | 2018 |
● I-Scan-2 ● Integro differential operator ● CHT ● Rubber sheet normalization ● Iridology chart ● GLCM ● Filter based feature selection method (fisher-score discrimination, t-test, chi-square test) ● Classifiers (BT, SVM, AB, GL, NN, RF) |
● Centre point and radius of pupil and iris ● Statistical, texture and discrete wavelength |
● Type 2 - diabetes | Accuracy: 89.66% |
| 21 | [77] | 2018 |
● I-Scan-2 ● Integro differential operator ● Rubber sheet normalization ● 2D-DWT ● Five classifiers (BT, RF, AB, SVM, NN) |
● Accuracy ● Specificity ● Sensitivity |
● Diabetes |
Accuracy: 59.63% Specificity: 96.87% Sensitivity: 98.8% |
| 22 | [78] | 2019 |
● ML based classification method (DT classifiers, SVM, ensemble classifiers) ● Iris segmentation ● Rubber sheet normalization ● Modified T-test ● PCA |
● Accuracy ● Sensitivity ● Specificity ● Precision ● F-score ● AUC |
● Type 2- diabetes | Accuracy: More than 95% |
| Breast disease | ||||||
| 23 | [98] | 2020 |
● Segmentation methods ● Watershed method ● Clustering techniques ● Graph based techniques ● Classifier techniques ● Morphology techniques ● Hybrid techniques |
● Evaluation metrics | ● Breast disease | - |
| 24 | [60] | 2021 |
● Genetic based artificial bee colony algorithm ● Ensemble classifiers (SVM, RF, DT, Naïve bayes, bagging, boosting) |
● Optimization parameters ● Cost based functions ● Fitness value ● Modification rate ● Recursive feature elimination |
● Chest pain | Accuracy: More than 90% |
| Covid 19 | ||||||
| 25 | [53] | 2021 |
● ML (supervised and unsupervised) ● Data fusion |
- | ● Covid-19 |
Accuracy with supervised ML: 92% Accuracy with unsupervised ML: 7.1% |
| 26 | [72] | 2021 |
● RNN ● CNN ● Hybrid DL model |
● Cough voice samples ● Blood samples ● Temperature |
● Covid-19 |
Accuracy with CT scan images: Above 94% Accuracy with x-ray images: Between 90-98% |
| 27 | [73] | 2021 |
● Nucleic acid-based ● Serological techniques |
- | ● Covid-19 | - |
| 28 | [36] | 2021 | ● Gradient-boosting machine model with DT base-learners |
● Cough ● Fever ● 60 + age ● Headache ● Sore throat ● Shortness of breath |
● Covid-19 | Accuracy: Above 80% |
| Heart disease | ||||||
| 29 | [18] | 2020 | ● DL algorithms |
● Sensitivity ● Specificity |
● Heart disease | Accuracy: 82% |
| 30 | [90] | 2022 |
● CNN ● Normalization |
● Mean absolute deviation ● Sensitivity ● Specificity |
● Cardiovascular disease | Median quality score: 19.6% |
| 31 | [48] | 2022 |
Metaheuristics optimization-based features selection algorithms: ● SALP swarm optimization algorithm ● Emperor penguin optimization algorithm ● Tree growth optimization algorithm |
● Aortic stenosis ● Mitral stenosis ● Mitral valve prolapses ● Mitral regurgitation |
● Valvular heart diseases |
Accuracy: Five classes: 98.53% Four classes: 98.84% Three classes: 99.07% Two classes: 99.70% |
| 32 | [59] | 2022 |
● Multifiltering ● REP tree ● M5P tree ● Random tree ● LR ● Naïve bayes ● J48 ● Jrip |
● Age ● Chest pain ● Blood pressure ● Cholesterol ● Fasting blood sugar ● Heart rate ● Slope ● ST depression ● Thalassemia |
● Cardiovascular disease |
Accuracy: 100% Lowest MAE: 0.0011 Lowest RMSE: 0.0231 Prediction time: 0.01 s |
| Respiratory disease | ||||||
| 33 | [6] | 2020 |
● DL ● Hilbert-huang transform |
● Multichannel lung sounds using statistical features of frequency modulations | ● Chronic obstructive pulmonary disease |
Accuracy: 93.67% Sensitivity: 91% Specificity: 96.33% |
| 34 | [43] | 2021 |
● AI ● ML |
- |
● Pulmonary function tests ● Diagnosis of a range of obstructive and restrictive lung diseases |
- |
| 35 | [69] | 2020 | ● CNN-MOE |
Audio recordings: ● Crackle ● Wheeze ● Crackle and wheeze ● Normal ● Time labels (onset and offset) |
Respiratory disease |
Accuracy: 4-class: 80% 3-class: 91% 2-class: 86-90% |
| 36 | [57] | 2019 | ● Improved bi-resnet DL architecture | ● Annotated respiratory cycles | Respiratory disease | Accuracy: 50.16% |
| Other | ||||||
| 37 | [7] | 2015 |
● TL ● Segmentation through voxel wise classification ● MRI scanners |
● MRI brain-segments ● White matter, gray matter, and cerebrospinal fluid segmentation ● Lesion segmentation |
- | Minimized classification error: 60% |
| 38 | [79] | 2018 |
● ML pipelining ● SVM classifier ● 5-fold cross-validation ● CMR imaging |
● Baseline left ventricular ● Ejection fraction ● Left ventricular circumferential strain ● Pulmonary regurgitation |
- |
Minor deterioration: 82% Major deterioration: 77% |
| 39 | [27] | 2019 |
● Image segmentation ● Feature selection ● Radiomic analysis ● Semantic analysis ● Lesion classification ● Pacs-side algorithm |
● Weighted sum ● Feature map |
- | - |
| 40 | [26] | 2020 |
● Data mining ● Pattern classification ● Neural nets ● CNN ● Lenet5 ● Max pooling |
● Feature extraction ● IRIS manipulation using SVM techniques |
- |
Accuracy: SVM: 82% CNN: 93.57% |
CNN Convolutional Neural Network, SVM Support Vector Machine, ML Machine Learning, DL Deep Learning, MRI Magnetic Resonance Imaging, PCA Principal Component Analysis, BT Binary Tree, RF Random Forest, NN Neural Network, AB Adaptive Boosting, CAD Computer-aided Diagnosis System, ANN Artificial Neural Network, AUC Area Under Curve, RMSE Root Mean Square Error, 2D-DWT Two-Dimensional Discrete Wavelet Transform, MAE Mean Absolute Error, QSSIM Quaternion Structure Similarity Index Metric, SSIM Structure Similarity Index Metric, PCC Pearson Correlation Coefficient, MoE Mixture of Experts, MIL Multiple Instance Learning, PPV Positive Predictive Value, NPV Negative Predictive Value, MCC Matthew’s Correlation Coefficient, TL Transfer Learning.