Establishment and validation of a multimodal model using deep learning for the recognition of early colorectal cancer

Table 1 Basic information of training set and validation set, n (%)

Item	Training set	Validation set	Statistical value	P value
Sample size	404	172	-	-
Age (years, mean ± SD)	61.17 ± 10.47	59.87 ± 10.21	t = 1.376	0.169
Lesion length (cm, mean ± SD)	1.23 ± 0.81	1.13 ± 0.73	t = 1.249	0.212
Gender			χ2 = 1.180	0.277
Female	141 (34.9)	52 (30.2)
Male	263 (65.1)	120 (69.8)
Pathological diagnosis			χ2 = 0.292	0.589
Non-cancerous	275 (68.1)	121 (70.3)
Early cancer	129 (31.9)	51 (29.7)
Lesion location			χ2 < 0.001	0.983
Colon	345 (85.4)	147 (85.5)
Rectum	59 (14.6)	25 (14.5)

Table 2 Comparison between non-cancerous and early colorectal cancer in the training set, n (%)

Item	Non-cancerous lesions	Early-stage cancer	Statistical value	P value
Sample size	275	129	-	-
Age (years, mean ± SD)	60.82 ± 10.85	61.91 ± 9.60	t = -0.972	0.332
Lesion length (cm, mean ± SD)	0.88 ± 0.46	1.95 ± 0.92	t = -15.630	< 0.001
Gender			χ2 = 1.264	0.261
Female	101 (36.7)	40 (31.0)
Male	174 (63.3)	89 (69.0)
Lesion location			χ2 = 15.817	< 0.001
Colon	248 (90.2)	97 (75.2)
Rectum	27 (9.8)	32 (24.8)

Table 3 Test performance of VGG19 model, ResNet50 model and InceptionV3 model in the training set

Model	Sensitivity	Specificity	Accuracy	AUROC
VGG19	0.78	0.93	0.88	0.92
ResNet50	0.57	0.94	0.82	0.84
InceptionV3	0.67	0.99	0.89	0.95
Ensemble deep learning prediction	0.85	0.93	0.90	0.89

AUROC: Area under the receiver operating characteristic curve.

Table 4 Diagnostic performance of the multimodal model, junior doctors, and senior doctors in the validation set

Regimentation	Sensitivity	Specificity	Accuracy	AUROC
The VGG19 model	0.43	0.82	0.70	0.75
The ResNet50 model	0.35	0.94	0.76	0.86
The InceptionV3 model	0.71	0.88	0.83	0.88
Clinical parameter group	0.53	0.94	0.81	0.88
Ensemble deep learning prediction	0.75	0.88	0.84	0.82
Junior endoscopist group	0.76	0.73	0.72	0.77
Senior endoscopist group	0.78	0.85	0.83	0.82
Multimodal model group	0.80	0.94	0.90	0.95

AUROC: Area under the receiver operating characteristic curve.