Table 3 Comparison of three models (23, 27, 31) with the same model but different parameters of replication, which vary image size.

From: Development of a novel deep learning method that transforms tabular input variables into images for the prediction of SLD

Model number	Model type	Vars	Image size	DA CN(RW/CL/NPP/NNP)/ F/R	Accuracy	Sensitivity	Specificity	Recall macro	AUC ROC	TN	FP	FN	TP	FP + FN
23	OursDLM	8	288 × 180	No/No/No	0.7917	0.8616	0.7664	0.8140	0.8562	338	103	22	137	125
–	OursDLM	8	288 × 50	No/No/No	0.6967	0.8616	0.6372	0.7494	0.8484	281	160	22	137	182
–	OursDLM	8	288 × 100	No/No/No	0.7683	0.8616	0.7347	0.7982	0.8476	324	117	22	137	139
–	OursDLM	8	288 × 210	No/No/No	0.7667	0.8616	0.7324	0.7970	0.8556	323	118	22	137	140
–	OursDLM	8	8 × 180	No/No/No	0.6117	0.8616	0.5215	0.6916	0.7670	230	211	22	137	233
–	OursDLM	8	40 × 180	No/No/No	0.6300	0.8616	0.5465	0.7041	0.7964	241	200	22	137	222
–	OursDLM	8	96 × 180	No/No/No	0.6800	0.8616	0.6145	0.7381	0.8292	271	170	22	137	192
–	OursDLM	8	192 × 180	No/No/No	0.7317	0.8616	0.6848	0.7732	0.8518	302	139	22	137	161
–	OursDLM	8	320 × 180	No/No/No	0.7217	0.8616	0.6712	0.7664	0.8540	296	145	22	137	167
27	OursDLM	8	288 × 180	(2/0/2/1)/No/No	0.8000	0.8365	0.7868	0.8117	0.8630	347	94	26	133	120
–	OursDLM	8	288 × 50	(2/0/2/1)/No/No	0.7717	0.8365	0.7483	0.7924	0.8600	330	111	26	133	137
–	OursDLM	8	288 × 100	(2/0/2/1)/No/No	0.7450	0.8365	0.7120	0.7742	0.8537	314	127	26	133	153
–	OursDLM	8	288 × 210	(2/0/2/1)/No/No	0.7883	0.8365	0.7710	0.8037	0.8632	340	101	26	133	127
–	OursDLM	8	8 × 180	(2/0/2/1)/No/No	0.5583	0.8365	0.4580	0.6473	0.7409	202	239	26	133	265
–	OursDLM	8	40 × 180	(2/0/2/1)/No/No	0.6633	0.8365	0.6009	0.7187	0.7963	265	176	26	133	202
–	OursDLM	8	96 × 180	(2/0/2/1)/No/No	0.7067	0.8365	0.6599	0.7482	0.8274	291	150	26	133	176
–	OursDLM	8	192 × 180	(2/0/2/1)/No/No	0.7083	0.8365	0.6621	0.7493	0.8339	292	149	26	133	175
–	OursDLM	8	320 × 180	(2/0/2/1)/No/No	0.7450	0.8365	0.7120	0.7742	0.8484	314	127	26	133	153
31	OursDLM	8	288 × 180	No/No/No	0.8033	0.8113	0.8004	0.8059	0.8562	353	88	30	129	118
–	OursDLM	8	288 × 50	No/No/No	0.7750	0.8113	0.7619	0.7866	0.8482	336	105	30	129	135
–	OursDLM	8	288 × 100	No/No/No	0.7867	0.8113	0.7778	0.7945	0.8469	343	98	30	129	128
–	OursDLM	8	288 × 210	No/No/No	0.7983	0.8113	0.7937	0.8025	0.8562	350	91	30	129	121
–	OursDLM	8	8 × 180	No/No/No	0.6550	0.8113	0.5986	0.7050	0.7670	264	177	30	129	207
–	OursDLM	8	40 × 180	No/No/No	0.6717	0.8113	0.6213	0.7163	0.7977	274	167	30	129	197
–	OursDLM	8	96 × 180	No/No/No	0.7200	0.8113	0.6871	0.7492	0.8252	303	138	30	129	168
–	OursDLM	8	192 × 180	No/No/No	0.7733	0.8113	0.7596	0.7855	0.8512	335	106	30	129	136
–	OursDLM	8	320 × 180	No/No/No	0.7650	0.8113	0.7483	0.7798	0.8542	330	111	30	129	141

CN, controlled noise; NPP, new positive patients created; NNP, new negative patients created; F, use of vertical flip; R, use of random rotation; RW, percentual variation to rows; CL, percentual variation to columns.
Best value for sensitivity level is shown in bold.

Back to article page

Table 3 Comparison of three models (23, 27, 31) with the same model but different parameters of replication, which vary image size.

Search

Quick links