Table 3.

Classification report for testing of transfer learning in HER2 scoring using VGG19 architecture for 100 epochs

Confusion matrix for the pre-trained VGG19 network with added FC layers
VGG19—patch-based scoring										VGG19 – Image based scoring with voting
			Predicted Score										Predicted score
			0/1+		2+		3+						0/1+		2+		3+
Ground truth	0/1+		298 (93.1%)		22 (6.9%)		0 (0.0%)			Ground truth	0/1+		40 (100.0%)		0 (0.0%)		0 (0.0%)
	2+		19 (7.9%)		221 (92.1%)		0 (0.0%)				2+		1 (3.3%)		29 (96.7%)		0 (0.0%)
	3+		5 (2.1%)		14 (5.8%)		221 (92.1%)				3+		0 (0.0%)		1 (3.3%)		29 (96.7%)
Classification report for the pre-trained VGG19 network with added FC layers
VGG19—patch-based scoring										VGG19—image-based scoring with voting
Class		Precision		Recall		F1-score		Support		Class		Precision		Recall		F1-Score		Support
0/1+		0.93		0.93		0.93		320		0/1+		0.98		1.00		0.99		40
2+		0.86		0.92		0.89		240		2+		0.97		0.97		0.97		30
3+		1.00		0.92		0.96		240		3+		1.00		0.97		0.98		30
Accuracy						0.93		800								0.98		100
Macro accuracy		0.93		0.92		0.93		800		Macro accuracy		0.98		0.98		0.98		100
Weighted accuracy		0.93		0.93		0.93		800		Weighted accuracy		0.98		0.98		0.98		100