Convolutional neural network-based vocal cord tumor classification technique for home-based self-prescreening purpose

Table 1 Results of the confusion matrix analysis of the four convolutional neural network models for each tumor type

Tumor	Model	TP	FP	FN	TN	Acc	Pre	Rec	Spe	F1
Cyst	Mask-50	28	19	25	299	0.8814	0.5957	0.5283	0.9403	0.5600
	Mask-101	25	24	30	291	0.8541	0.5102	0.4545	0.9238	0.4808
	Yolo-4	41	11	14	302	0.9321	0.7885	0.7455	0.9649	0.7664
	SSD-MN	24	11	30	301	0.8880	0.6857	0.4444	0.9647	0.5393
Granuloma	Mask-50	85	34	2	262	0.9060	0.7143	0.9770	0.8851	0.8252
	Mask-101	74	11	8	278	0.9488	0.8706	0.9024	0.9619	0.8862
	Yolo-4	79	1	1	287	0.9946	0.9875	0.9875	0.9965	0.9875
	SSD-MN	70	6	8	281	0.9616	0.9211	0.8974	0.9791	0.9091
Leukoplakia	Mask-50	49	36	21	277	0.8512	0.5765	0.7000	0.8850	0.6323
	Mask-101	92	132	7	226	0.6958	0.4107	0.9293	0.6316	0.5697
	Yolo-4	46	6	14	303	0.9458	0.8846	0.7667	0.9806	0.8214
	SSD-MN	42	6	21	302	0.9272	0.8750	0.6667	0.9805	0.7568
Nodule	Mask-50	51	27	14	296	0.8943	0.6538	0.7846	0.9164	0.7133
	Mask-101	61	119	14	246	0.6977	0.3387	0.8133	0.6740	0.4784
	Yolo-4	41	2	17	313	0.9491	0.9535	0.7069	0.9937	0.8119
	SSD-MN	33	13	24	306	0.9016	0.7174	0.5789	0.9592	0.6408
Polyp	Mask-50	146	65	18	190	0.8019	0.6919	0.8902	0.7451	0.7787
	Mask-101	105	32	36	210	0.8225	0.7664	0.7447	0.8678	0.7554
	Yolo-4	110	22	24	218	0.8770	0.8333	0.8209	0.9083	0.8271
	SSD-MN	64	9	68	229	0.7919	0.8767	0.4848	0.9622	0.6244

Bold values in the table represent the cases of the lowest error (lowest in false negative) and the best performance (highest in F1-score) among the four models
TP true-positive, FP false-positive, FN false-negative, TN true-negative, Acc accuracy, Pre precision, Rec recall, Spe specificity, F1 F1-score

ISSN: 1475-925X