IRR-Net: A Joint Learning Framework for Image Reconstruction and Recognition of Photoacoustic Tomography

Table 8

Evaluation metrics for multiclass classification of in vivo images.

Method		Acc							F1 score	Runtime (s)
		Mandible	Tongue	Liver	Kidney	Spinal cord	Abdom aorta	Rib
HH (5 DGD + GoogLeNet)		0.8134	0.8067	0.8203	0.7946	0.8286	0.7798	0.7802	0.6251	24.1
IRR-Net		0.7941	0.7855	0.7995	0.7667	0.8109	0.7657	0.7575	0.6135	12.0
HH (DGD + U-Net+ResNet)		0.7941	0.7855	0.7995	0.7667	0.8109	0.7657	0.7575	0.6135	33.7
LH	1 DGD + GoogLeNet	0.7884	0.7811	0.7805	0.7602	0.7968	0.7613	0.7536	0.5904	10.9
	TR + GoogLeNet	0.7733	0.7766	0.7741	0.7545	0.7839	0.7574	0.7504	0.5745	3.1
	TR + AlexNet	0.7686	0.7705	0.7687	0.7478	0.7743	0.7536	0.7488	0.5544	2.7
	TR + ResNet	0.7677	0.7700	0.7583	0.7443	0.7725	0.7524	0.7465	0.5537	2.7
HL (5 DGD + SVM)		0.7670	0.7698	0.7452	0.7401	0.7708	0.7515	0.7440	0.5531	21.5
LL (TR + SVM)		0.7560	0.7681	0.7227	0.7354	0.7686	0.7503	0.7380	0.5433	1.5