Table 2 The top two models for different sensitivity levels, with sensitivity ≥ 0.8052, our DL model, and our traditional ML model (model number 1 (T1) to 32 (T32), and up to 14 variables.

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
1	OursDLM	11	110 × 300	(1/0/5/5)/No/No	0.5800	1	0.4286	0.7143	0.8163	189	252	0	159	252
T1	BRF	13	–	(0.1/-/5/5)/-/-	0.5367	1	0.3696	0.6848	0.8682	163	278	0	159	278
3	OursDLM	8	288 × 210	(5/0/5/5)/No/No	0.6533	0.9874	0.5329	0.7602	0.8577	235	206	2	157	208
T3	MLP	7	–	No	0.6017	0.9874	0.4626	0.7250	0.8639	204	237	2	157	239
5	OursDLM	8	288 × 180	(0.1/0.1/5/5)/No/No	0.7083	0.9748	0.6122	0.7935	0.8637	270	171	4	155	175
T5	MLP	10	–	(1/-/5/5)/-/-	0.6367	0.9748	0.5147	0.7448	0.8669	227	214	4	155	218
7	OursDLM	8	288 × 180	(0.01/0/5/5)/No/No	0.7117	0.9623	0.6213	0.7918	0.8623	274	167	6	153	173
T7	BRF	14	-	No	0.6950	0.9623	0.5986	0.7805	0.8718	264	177	6	153	183
9	OursDLM	8	288 × 180	(1.5/1.5/5/5)/No/Yes	0.7233	0.9497	0.6417	0.7957	0.8662	283	158	8	151	166
T9	BRF	11	–	No	0.7033	0.9497	0.6145	0.7821	0.8721	271	170	8	151	178
11	OursDLM	8	288 × 180	(2/0/2/0)/No/No	0.7367	0.9371	0.6644	0.8008	0.8587	293	148	10	149	158
T11	SVM	11	–	(0.1/-/10/10)/-/-	0.7100	0.9371	0.6281	0.7826	0.8640	277	164	10	149	174
13	OursDLM	8	288 × 180	(5/5/1/0)/No/Yes	0.7433	0.9245	0.6780	0.8013	0.8681	299	142	12	147	154
T13	SVM	10	–	(2/-/10/10)/-/-	0.7183	0.9245	0.6440	0.7843	0.8678	284	157	12	147	169
15	OursDLM	8	288 × 180	(2/0/1/0)/No/No	0.7500	0.9120	0.6916	0.8018	0.8588	305	136	14	145	150
T15	SVM	9	–	(5/-/10/10)/-/-	0.7267	0.9120	0.6599	0.7859	0.8621	291	150	14	145	164
17	OursDLM	8	288 × 180	(0.01/0/2/0)/No/No	0.7683	0.8994	0.7211	0.8102	0.8660	318	123	16	143	139
T17	MLP	11	–	(1.5/-/5/5)/-/-	0.7367	0.8994	0.6780	0.7887	0.8634	299	142	16	143	158
18	OursDLM	8	288 × 180	(1/0/1/1)/No/No	0.7700	0.8931	0.7256	0.8094	0.8667	320	121	17	142	138
T18	SVM	13	–	(0.01/-/10/10)/-/-	0.7417	0.8931	0.6871	0.7901	0.8552	303	138	17	142	155
19	OursDLM	8	288 × 180	(5/0.01/5/5)/No/No	0.7750	0.8868	0.7347	0.8107	0.8664	324	117	18	141	135
T19	MLP	10	-	(1.5/-/5/5)/-/-	0.7500	0.8868	0.7001	0.7937	0.8592	309	132	18	141	150
22	OursDLM	8	288 × 180	(5/0/5/5)/No/No	0.7833	0.8679	0.7528	0.8104	0.8666	332	109	21	138	130
T22	SVM	11	–	(0.1/-/10/10)/-/-	0.7683	0.8679	0.7324	0.8002	0.8619	323	118	21	138	139
23	OursDLM	8	288 × 180	No/No/No	0.7917	0.8616	0.7664	0.8140	0.8562	338	103	22	137	125
T23	SVM	11	-	(0.1/-/10/10)/-/-	0.7683	0.8616	0.7347	0.7982	0.8619	324	117	22	137	139
25	OursDLM	8	288 × 180	(0/0.01/5/5)/No/No	0.7950	0.8491	0.7755	0.8123	0.8561	342	99	24	135	123
T25	SVM	11	–	(0.1/-/10/10)/-/-	0.7800	0.8491	0.7551	0.8021	0.8619	333	108	24	135	132
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
T27	MLP	11	–	(2/-/10/10)/-/-	0.7767	0.8365	0.7551	0.7958	0.8601	333	108	26	133	134
29	OursDLM	8	288 × 180	(0.1/0/1/0)/No/No	0.8000	0.8239	0.7914	0.8076	0.8561	349	92	28	131	120
T29	MLP	13	–	(2/-/10/10)/-/-	0.7933	0.8239	0.7823	0.8031	0.8680	345	96	28	131	124
31	OursDLM	8	288 × 180	No/No/No	0.8033	0.8113	0.8004	0.8059	0.8562	353	88	30	129	118
T31	MLP	14	–	(1/-/5/5)/-/-	0.7950	0.8113	0.7891	0.8002	0.8555	348	93	30	129	123
32	OursDLM	8	288 × 180	(5/0/1/0)/No/No	0.8050	0.8050	0.8050	0.8050	0.8645	355	86	31	128	117
T32	MLP	8	–	(1.5/-/5/5)/-/-	0.8000	0.8050	0.7982	0.8016	0.8587	352	89	31	128	120

The main hyperparameters are: number of variables, image size for the DL model and the type of DA. The accuracy, sensitivity, specificity, recall macro, AUCROC, TN, FP, FN, TP, and FP + FN are also shown.
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 2 The top two models for different sensitivity levels, with sensitivity ≥ 0.8052, our DL model, and our traditional ML model (model number 1 (T1) to 32 (T32), and up to 14 variables.

Search

Quick links