Table 1. A summary of recent related studies.

Article	Purpose	Class	Method	Accuracy (%)	Limitation
Khan et al. (2020b)	Ulcer, polyp, bleeding detection	4	RCNN, ResNet101, and SVM	99.13	Used only spatial features. Low segmentation accuracy for the ulcer regions. Fail for the segmentation of polyp and bleeding regions.
Khan et al. (2020a)	Ulcer, and bleeding detection	3	VGG-16, PSO, and SVM	98.4	Limited classes Used only spatial features. High computational cost
Igarashi et al. (2020)	Classify several GI diseases	14	AlexNet	96.5	Used only spatial features. The training or test data included chosen images of gastric cancer lesions, which could cause a selection bias. Has high computational cost Cannot be used in real-time examinations
Alaskar et al. (2019)	Ulcer detection	2	AlexNet & Google Net	97.143	Limited classes. Used only spatial features
Owais et al. (2019)	Classification of multiple GI diseases	37	ResNet-18 and LSTM	89.95	High computational cost. Used individual type of features Low accuracy
Fan et al. (2018)	Ulcer and Erosion detection	2	AlexNet	95.16 95.34	Limited classes. Used only spatial features. Used only one type of CNN features The CADx was applied separately for ulcer and Erosion detection
He et al. (2018)	Hookworm detection	2	VGG-16 and Inception	88.5	Limited classes. Used only spatial features Low accuracy
Yuan & Meng (2017)	Polyps detection	2	Stacked sparse auto-encoder with image manifold	98	Limited classes. Used only spatial features
Pei et al. (2017)	Bowel detection and assessment	2	LSTM and PCA	88.8	Limited classes. Used only temporal features. Used only one type of CNN features Low accuracy Small dataset
Sharif et al. (2019)	Ulcer, and bleeding detection	3	VGG-16, VGG-19, geometric features, KNN	99.42	Limited classes. Small dataset. Used spatial and geometric features only
Ghatwary, Ye & Zolgharni (2019)	Esophageal cancer detection	2	Gabor Filter. faster R-CNN, and SVM	95	Limited classes. Used only one type of CNN features Used spatial and textural based -Gabor features only. High computational cost
Billah, Waheed & Rahman (2017)	Polyps detection	2	Color based DWT,CNN, and SVM	98.65	Limited classes. Used only one type of CNN features Used spatial and color based –DWT only Small dataset
Nadeem et al. (2018)	Classification of several GI diseases	8	VGG-19, Haralick and LBP texture analysis, and logistic regression	83	Low accuracy Used only one type of CNN features Used spatial features based on CNN and textural analysis only
Majid et al. (2020)	Bleeding, esophagitis, polyp, and ulcerative-colitis classification	5	DCT, color based statistical features, DWT, VGG-16, GA, and E	96.5	High computational cost. Used only one type of CNN DL features
Nguyen et al. (2020)	Classifying images to normal and abnormal	2	DenseNet, Inception, and VGG-16	70.7	Classify images to either normal or abnormal. Did not classify several GI diseases. Low accuracy
Owais et al. (2020)	Classification of multiple GI diseases	37	DenseNet and LSTM	95.75	High computational cost. Used individual type of features