Deep learning predicts cervical lymph node metastasis in clinically node-negative papillary thyroid carcinoma

Insights into Imaging

Table 3 Confusion matrices of the four models and three radiologists according to the test sets

a. Confusion matrices of DCNN models on test set A
Prediction	BMUS (truth)		CDFI (truth)		Clinical (truth)		Ensemble (truth)
Prediction	Non-metastasis	Metastasis	Non-metastasis	Metastasis	Non-metastasis	Metastasis	Non-metastasis	Metastasis
Non-metastasis	34	9	26	7	30	18	35	8
Metastasis	13	38	21	40	17	29	12	39
b. Confusion matrices of radiologists on test set A
Prediction	Expert 1 (truth)		Expert 2 (truth)		Expert 3 (truth)
Prediction	Non-metastasis	Metastasis	Non-metastasis	Metastasis	Non-metastasis	Metastasis
Non-metastasis	30	18	32	27	24	25
Metastasis	17	29	15	20	23	22
c. Confusion matrices of DCNN models on test set B
Prediction	BMUS (truth)		CDFI (truth)		Clinical (truth)		Ensemble (truth)
Prediction	Non-metastasis	Metastasis	Non-metastasis	Metastasis	Non-metastasis	Metastasis	Non-metastasis	Metastasis
Non-metastasis	27	12	26	10	26	15	31	11
Metastasis	18	32	19	34	19	29	14	33
d. Confusion matrices of radiologists on test set B
Prediction	Expert 1 (truth)		Expert 2 (truth)		Expert 3 (truth)
Prediction	Non-metastasis	Metastasis	Non-metastasis	Metastasis	Non-metastasis	Metastasis
Non-metastasis	29	15	24	17	26	25
Metastasis	15	30	20	28	18	20