Fine-Grained Lung Cancer Classification from PET and CT Images Based on Multidimensional Attention Mechanism

Table 3

Evaluation of different fusion strategies.

Network architecture	Data	AUC				Balanced accuracy
		Score of each category		Average score	Variance
MFSCA-DenseNet (GMU)	CT + PET	LUSC	0.843	0.840	0.07	0.67
		LUAD	0.899
		SCLC	0.868
MFSCA-DenseNet (GMU + joint optimization)	CT + PET	LUSC	0.938	0.920	0.05	0.72
		LUAD	0.910
		SCLC	0.913
MFSCA-DenseNet (connection)	CT + PET	LUSC	0.771	0.801	0.12	0.63
		LUAD	0.703
		SCLC	0.885
MFSCA-DenseNet (connection + joint optimization)	CT + PET	LUSC	0.788	0.815	0.10	0.65
		LUAD	0.759
		SCLC	0.897