Study Dataset (Population) Method Measures Findings
Bai et al. (2018) [127]
N = 4875 UK Biobank
FCN
Mean absolute distance, Hausdorff distance
Human-like performance in evaluating CMR images. Dice metric= 0.94 (LV cavity), 0.88 (LV myocardium) 0.90 (RV cavity).
Wang et al. (2017) [128] N = 840 (210) mammograms CNN:10 convolution layers and two fully connected layers Free response receiver operating characteristics (FROC) analysis. Two-class classification problem to categorize breast arterial calcification (BAC) pixels and non-BAC pixels in mammograms, a risk marker of CAD. Calcium mass quantification analysis
Jamthikar et al. (2020) [129]
N = 404 (202)
ML-based risk factor classifier AUC = 0.99 (P < 0.001) compared to conventional 57.14% improvement over conventional algorithm. PoM and FoM = 96%. Mean absolute error < 5%.
Biswas et al. (2018) [130] N = 63, Liver US: 36 with FLD, 27 normal CNN, SVM, ELM. Tissue characterization, risk stratification
ROC, reliability index, timing analysis.
DL: Robustness to noise, 100% accuracy at 15% cropping of borders. DL better than SVM & ELM
Kuppili et al. [131]
N = 63
ELM, SVM
AUC, speed test validation 96.75% accuracy in ELM, 89.01% accuracy in SVM AUC 0.97 and 0.91 in ELM and SVM, respectively. Speed improvement of 40% in ELM
Hemalatha et al. [132]
N = 276 MEDUSA database Hit-or-miss transform for bone-line segmentation, Active contour for localization Accuracy, precision, specificity, sensitivity, ROC \(95.02\%\ \pm 2.78\) accuracy Increase in true positives from 78.12% to 98.15%. False positives decreased by 1.41%. CNN applied for classification
Poplin et al. [133]
N = 48,101, N = 12,026 Validation: UK Biobank N = 236,234, N = 999 EyePACS
Inception-v3 neural network architecture.
MAE in prediction. AUC for the binary classifier. Cohen’s kappa for multiclass classification Age prediction: MAE 3.26 years and 3.48 in UK Biobank and EyePACS. Ethnicity prediction: Kappa score of 0.6 and 0.75. Prediction on the onset of MACE: AUC of 0.7 from retinal fundus images alone