Table 4 Performance metrics of deep learning algorithms (CNN-Base, VGG16, ResNet101, DenseNet121, inception v3, and ResNet50) across two test sets (T1, T2) for classifying four severity levels of corneal pathology: healthy (H), mild (M), moderate (MO), and severe (S). The performance metrics include the accuracy, sensitivity (Sn), specificity (Sp), F1 score, hamming distance (HD), receiver operating characteristic (ROC)-AUC (area under the curve), and precision recall-area under the curve (PR-AUC). Micro- and Macroaverages for the ROC-AUC and PR-AUC are provided for overall model evaluation.
From: Artificial intelligence derived grading of mustard gas induced corneal injury and opacity
Model | Performance metrics | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Accuracy | Sn | Sp | F1 | HD (Avg.) | ROC-AUC | PR-AUC | |||||||||
Train | T1 | T2 | Class | T1 | T2 | T1 | T2 | T1 | T2 | Micro | Macro | Micro | Macro | ||
Baseline-CNN | 0.64 | 0.70 | 0.63 | H | 0.73 | 0.87 | 0.93 | 0.82 | 0.76 | 0.67 | 0.40 | 0.88 | 0.90 | 0.65 | 0.72 |
M | 0.75 | 0.43 | 0.97 | 0.89 | 0.80 | 0.52 | |||||||||
MO | 0.50 | 0.40 | 0.79 | 0.84 | 0.50 | 0.42 | |||||||||
S | 0.60 | 0.67 | 0.81 | 0.88 | 0.55 | 0.63 | |||||||||
VGG16 | 0.75 | 0.80 | 0.74 | H | 0.80 | 0.83 | 0.85 | 0.91 | 0.76 | 0.81 | 0.29 | 0.93 | 0.93 | 0.82 | 0.80 |
M | 0.60 | 057 | 0.87 | 0.90 | 0.60 | 0.59 | |||||||||
MO | 0.67 | 0.50 | 0.96 | 0.95 | 0.74 | 0.62 | |||||||||
S | 0.91 | 0.92 | 0.96 | 0.84 | 0.87 | 0.71 | |||||||||
ResNet 101 | 0.75 | 0.77 | 0.80 | H | 0.84 | 0.86 | 0.93 | 0.90 | 0.84 | 0.77 | 0.29 | 0.92 | 0.93 | 0.79 | 0.80 |
M | 0.45 | 0.67 | 0.90 | 0.88 | 0.48 | 0.70 | |||||||||
MO | 0.56 | 0.69 | 0.84 | 0.92 | 0.57 | 0.69 | |||||||||
S | 0.77 | 0.79 | 0.90 | 0.96 | 0.71 | 0.81 | |||||||||
DenseNet121 | 0.80 | 0.81 | 0.85 | H | 0.89 | 0.92 | 0.95 | 0.91 | 0.89 | 0.84 | 0.20 | 0.94 | 0.97 | 0.84 | 0.89 |
M | 0.55 | 0.62 | 0.90 | 0.95 | 0.55 | 0.72 | |||||||||
MO | 0.72 | 1.00 | 0.91 | 0.88 | 0.74 | 0.81 | |||||||||
S | 0.92 | 0.79 | 0.96 | 1.00 | 0.89 | 0.88 | |||||||||
InceptionV3 | 0.77 | 0.73 | 0.77 | H | 0.79 | 0.89 | 0.88 | 0.88 | 0.77 | 0.83 | 0.28 | 0.90 | 0.91 | 0.76 | 0.78 |
M | 0.39 | 0.40 | 0.86 | 0.94 | 0.45 | 0.50 | |||||||||
MO | 0.69 | 0.67 | 0.92 | 0.89 | 0.69 | 0.67 | |||||||||
S | 1.00 | 0.92 | 0.92 | 0.92 | 0.85 | 0.83 | |||||||||
ResNet50 | 0.87 | 0.85 | 0.83 | H | 0.90 | 0.78 | 1.00 | 0.98 | 0.95 | 0.80 | 0.17 | 0.94 | 0.95 | 0.80 | 0.84 |
M | 0.88 | 0.79 | 0.96 | 0.87 | 0.88 | 0.79 | |||||||||
MO | 0.80 | 0.88 | 0.91 | 0.91 | 0.77 | 0.83 | |||||||||
S | 0.78 | 0.92 | 0.94 | 0.98 | 0.74 | 0.92 | |||||||||
