Table 2.

Detailed evaluation metrics for disc state classification of ROI images, comparing performance with and without fine-tuned models.

Models	Accuracy with fine-tuning (%)	Accuracy without fine-tuning (%)	AUC (%)	Specificity (%)	Precision (%)	Recall (%)	F1-score (%)
VGG16	89.87	81.01	82.98	100	100	65.96	79.49
VGG19	92.41	82.28	87.23	100	100	74.47	85.37
ResNet50	96.84	94.94	95.91	98.20	95.65	93.62	94.62
DenseNet121	93.04	89.24	88.91	99.09	97.37	78.73	87.06
InceptionV3	93.03	92.41	91.98	94.60	87.50	89.36	88.42
Xception	96.84	90.51	96.52	97.30	93.75	95.75	94.74
AEM	98.10	93.67	96.81	100	100	93.62	96.70
WAEM	99.37	96.84	98.94	100	100	97.87	98.93