Table 2.

Classification performance (%) of the examined deep neural network architectures following Approach 1 when using the pretrained weights for the base network and when training each network end-to-end.

Base Model	End-to-End Training						Pre-Trained Base
	Accuracy	F1-Score	Precision	Recall	Jaccard	Dice	Accuracy	F1-Score	Precision	Recall	Jaccard	Dice
EfficientNetB4	98.04	98.22	98.52	97.95	96.52	98.23	89.95	88.71	92.86	85.55	80.35	89.10
EfficientNetB7	96.87	96.66	97.10	96.27	93.24	96.50	88.67	85.70	93.11	81.46	78.35	87.86
VGG16	94.77	95.39	96.11	94.84	88.83	94.08	87.57	85.59	89.23	82.86	72.85	84.29
Xception	98.00	98.20	98.58	97.87	95.98	97.95	90.36	90.07	92.29	88.16	79.92	88.84
InceptionResNetV2	97.38	97.35	97.88	96.89	94.17	97.00	89.05	88.76	91.25	86.65	77.51	87.33
InceptionV3	96.18	95.76	96.73	95.07	90.63	95.09	88.98	88.30	91.13	85.90	76.03	86.39
MobileNetV2	96.90	97.00	97.52	96.53	93.09	96.42	89.40	88.44	92.11	85.57	78.66	88.06
ResNet50V2	97.45	97.92	98.20	97.67	94.94	97.41	91.02	90.81	93.43	88.68	81.23	89.64
DenseNet121	97.07	97.46	97.95	97.03	93.83	96.82	88.06	87.28	93.23	83.02	75.66	86.14
DenseNet169	97.49	97.68	97.76	97.63	94.82	97.34	89.71	88.72	92.60	85.64	79.10	88.33
DenseNet201	97.25	97.15	97.60	96.77	93.93	96.87	89.12	88.25	91.63	85.58	78.21	87.77

Note: Results in bold denote the best performance for each metric and approach. Underlined results denote the overall best performance for each metric.