Table 6 Results of Step-Wise Fine-Tuning for ResNet50⁺⁺ verses LoRA for ResNet50⁺⁺

Step	Epoch	Acc.¹	Precision			Recall			F1-Score		AUC
Step	Epoch	Acc.¹	Benign	Malignant	MA²	Benign	Malignant	MA³	Benign	Malignant	AUC
1	44	0.738	0.743	0.702	0.723	0.946	0.282	0.614	0.832	0.402	0.766
2	29	0.901	0.925	0.845	0.885	0.931	0.834	0.883	0.928	0.84	0.957
3	34	0.942	0.957	0.908	0.933	0.959	0.904	0.931	0.958	0.906	0.979
4	29	0.937	0.95	0.906	0.928	0.958	0.889	0.924	0.954	0.897	0.982
5	32	0.938	0.949	0.913	0.931	0.962	0.885	0.923	0.955	0.899	0.987
6	30	0.938	0.952	0.907	0.930	0.958	0.895	0.926	0.955	0.901	0.981
1 (LoRA)	40	0.736	0.748	0.670	0.709	0.931	0.309	0.620	0.829	0.423	0.771
2 (LoRA)	51	0.899	0.923	0.847	0.885	0.931	0.829	0.880	0.927	0.837	0.958
3 (LoRA)	41	0.921	0.941	0.877	0.909	0.944	0.870	0.907	0.943	0.873	0.974
4 (LoRA)	32	0.920	0.938	0.880	0.909	0.946	0.862	0.904	0.942	0.871	0.972
5 (LoRA)	41	0.925	0.946	0.880	0.913	0.945	0.881	0.913	0.946	0.880	0.978
6 (LoRA)	32	0.922	0.949	0.866	0.907	0.937	0.889	0.913	0.943	0.877	0.973

Quick links

Search