Ultrasound-based deep learning radiomics model for differentiating benign, borderline, and malignant ovarian tumours: a multi-class classification exploratory study

Table 3 The error analysis of the three-class classification prediction model

Signature	Real label	class	Prediction			sum
			Benign ovarian tumour	BOT	Malignant ovarian tumour
Clinic_Sig	Real label	Benign ovarian tumour	98(89.91%)	0(0%)	11(10.09%)	109
		BOT	11(100%)	0(0%)	0(0%)	11
		Malignant ovarian tumour	41(83.67%)	0(0%)	8(16.33%)	49
		sum	150	0	19	169
Rad_Sig	Real label	Benign ovarian tumour	97(88.99%)	0(0%)	12(11.01%)	109
		BOT	6(54.55%)	0(0%)	5(45.45%)	11
		Malignant ovarian tumour	30(61.22%)	0(0%)	19(38.78%)	49
		sum	133	0	36	169
DTL_Sig	Real label	Benign ovarian tumour	94(86.24%)	1(0.92%)	14(12.84%)	109
		BOT	4(36.36%)	4(36.36%)	3(27.28%)	11
		Malignant ovarian tumour	14(28.58%)	6(12.24%)	29(59.18%)	49
		sum	112	11	46	169
DRL_Sig	Real label	Benign ovarian tumour	90(82.57%)	2(1.83%)	17(15.60%)	109
		BOT	4(36.36%)	6(54.55%)	1(9.09%)	11
		Malignant ovarian tumour	12(24.49%)	6(12.24%)	31(63.27%)	49
		sum	106	14	49	169

Clinic_Sig: clinical signature; Rad_Sig: radiomics signature; DTL_Sig: deep transfer learning signature; DLR_Sig: deep learning radiomic signature; BOT: Borderline ovarian tumour

ISSN: 1471-2342