Table 3.

Studies using deep learning approach for bladder cancer treatment.

Author	Year	Modality	Number of patients (Train/Val/Test)	CNN structure	Performance (validation or testing dataset)
Cha et al. (25)	2016	CT	62, LOOCV	A network contains 2 convolution layers, 2 locally connected layers, and 1 fully connected layer.	AUC = 0.73
Cha et al. (66)	2017	CT	82	A network contains 2 convolution layers, 2 locally connected layers, and 1 fully connected layer. Each layer contains 16 kernals.	AUC = 0.73
Wu et al. (67)	2019	CT	73/9/41	The basic network contains 2 convolution layers, 2 locally connected layers, and 1 fully connected layer.	AUC (basic-random weights) = 0.73 AUC (basic-pretrained weights) = 0.79 AUC (DL-CNN-1) = 0.72 AUC (DL-CNN-2) = 0.86 AUC (DL-CNN-3) = 0.69 AUC (C1 Frozen) = 0.81 AUC (C1,C2 Frozen) = 0.78 AUC (C1,C2,L3 Frozen) = 0.71
Cha et al. (68)	2019	CT	123, LOOCV	DL-CNN with a radiomics assessment model	AUC (CDSS-T only) = 0.80 AUC (with CDSS-T) = 0.77 AUC (No CDSS-T) = 0.74

AUC, area under the curve; LOOCV, leave-one-out cross-validation.