Table 1 Summary of survey on recent GI tract bleeding classification methods.
Author | Features used | Segmentation method | Classification method | Dataset |
|---|---|---|---|---|
Pedro et al.16 | CNN features | MI-RCNN, PANet with mask subnet | Integrated with segmentation | KID |
Furqan Rustam et al.17 | MobileNet + custom CNN features | None | MobileNet + CNN | 1,650 WCE images |
Jain et al.18 | CNN features + attention maps | Grad-CAM++ + SegNet | WCENet CNN | KID |
Muhammad Attique Khan et al.19 | VGG16 transfer learning features + PSO selection | Saliency-based ulcer detection | Cubic SVM | Custom ulcer dataset |
Muhammad Sharif et al.20 | Geometric + CNN features (VGG16, VGG19) | Contrast-enhanced color feature extraction | KNN | 5,500 WCE images |
Samira Lafraxo et al.21 | CNN attention features | AttResU-Net | Integrated | - |
Goel et al.22 | Integrated features in CNN | CNN segmentation | Same CNN | KID (48 images) |
Vrushali Raut et al.5 | GLCM, SIFT, LBP, HOG + CNN | Modified U-Net + DH-DSU | Enhanced DNN | KID Atlas |
Bai et al.12 | CAM-based features | Discrepancy decoder + CAMPUS | Integrated | - |
Chathurika et al.23 | DenseNet201 + ResNet18 + VGG16 ensemble features | None | GAP + SVD | WCE dataset |
Marwa et al.24 | CapsNet features + CAL | None | DBN-ELM | GI bleeding dataset |
Rathnamala et al.25 | Enhanced Oriented FAST and Rotated BRIEF (E-ORB) keypoints | Adaptive Dense U-Net | Multi-Enhanced Deep CapsNet | Public WCE image dataset (source not specified) |
Sushma et al.26 | CNN deep features with Adaptive Attention Module | Attention-augmented CNN feature maps (implicit segmentation via attention) | Angular Contrastive (AC) loss-based CNN classifier | WCE bleeding frames dataset (collected) |
Singh et al.27 | DenseNet-121 deep features | U-Net | DenseNet feature embeddings with integrated classification head | Auto-WCBleedGen Challenge Dataset |
