Table 8.
Summary of DL-based studies for colorectal cancer diagnosis.
| Author (Year) | Topic | Imaging Modality | DL Architecture | Datasets Availability | Results |
|---|---|---|---|---|---|
| Hou et al. [50] (2019) | Polyp segmentation | HIs | CNN | Public | Reduce the error of segmentation by 7.8%, 5.4%, and 3.2%. |
| Janowczyk et al. [53] (2016) | Polyp Detection | Digital-Pathology (DP) | NIA | Public | TPR: 86% PPV: 64% |
| Tripathi et al. [113] (2020) | Polyp Detection | HIs | AlexNet VGG16 VGG19 ResNet50 DenseNet121 InceptionV3 |
Public | Precision: 0.62% Recall: 0.63% AUC: 0.03% Loss: 0.0043 |
| Shapcott et al. [101] (2019) | Polyp detection | HIs | CNNs | Private | Accuracy: 65% |
| Ben Hamida et al. [117] (2021) | Polyp detection | Digital pathology (DP) | ALEXNET | Public | CRC-5000-Accuracy: 98.66% NCT-CRC-HE-Accuracy: 99.12% |
| Liewa et al. [118] (2021) | Polyp Detection CVC-ClinicDB | Endoscopic Images | ResNet-50 | Public | Accuracy: 99.10% Sensitivity: 98.82% Precision: 99.37% Specificity: 99.38% |
| Pacal et al. [20] (2020) | Polyp Detection | CT | RNNs, Autoencoders (AEs) | Public | sensitivity: 0.91 |
| De et al. [103] (2019) | Polyp Segmentation | Colonoscopy Images | CNNs | Public | F1-Score: 91.4% FPR: 0.079 |
| Javed et al. [70] (2020) | Detection of Polyp | Colonoscopy Images | CNN | Public | Specificity: 920% Accuracy: 89.8% F1: 91.4% Recall: 89.8% Precision: 93.6% |
| Sikder et al. [119] (2021) | Polyp Detection | MRI | CNN | Private | Accuracy: 93% |
| Kang et al. [104] (2019) | Polyp Segmentation | Colonoscopy | CNN | Public | Dataset results of Etis-Larib: recall: 74.37%, precision: 73.84%, IoU: 66.07% |
| Sornapudi et al. [105] (2019) | Detection of Polyp | WCE + Colonoscopy | CNN | Public | Dataset results of ResNet-101: F2: 78.70% recall: 80.29%, F1: 76.43%, precision: 72.93%. Dataset results of ResNet-50: recall: 67.79%, F2: 66.57%, precision: 62.11%, F1: 64.83% |
| Jia et al. [112] (2020) | Segmentation of Polyp | Colonoscopy | CNN | Public | recall: 81.7%, Precision: 63.9%, F2: 77.4%, F1: 71.7% |
| Ozawa et al. [116] (2020) | Polyp classification of colorectal and automated detection of endoscopic | Colonoscopy | CNN | Private | Detection: PPV: 86%, sensitivity: 92% Narrow-band images classification: 81% Conventional white-light Classification images: 83% |
| Zhang et al. [106] (2019) | Detection of Polyps | Colonoscopy | CNN | Private | F1: 84.24%, recall: 76.37%, Precision: 93.92% |
| Zobel et al. [107] (2019) | Polyp Detection | Colonoscopy | CNN | Private | F1: 89%, precision: 86%, Recall: 93% |
| Ma et al. [108] (2019) | Polyp Localization | Colonoscopy | CNN | Private | sensitivity: 93.67%, accuracy: 96%, AP: 94.92%, specificity: 98.36% |
| Shaban et al. [64] (2020) | Polyp Detection and Classification | Colonoscopy | CNN | Private | F2: 66.07%, F1: 68.72% |
| Blanes-Vidal et al. [109] (2019) | Polyp detection | WCE | CNN | Private | sensitivity: 97.1%, Accuracy: 96.4%, specificity: 93.3% |
| Wang et al. [110] (2019) | Real-time automatic detection system | Colonoscopy | CNN | Private | ADR Increment 9.1% vs. 20.3%, p < 0.001) |
| Mostafiz et al. [115] (2020) | Polyp Detection | Colonoscopy | CNN + CEMD | Public | sensitivity: 99.91%, Accuracy: 99.53%, specificity: 99.15% |
| Yuan et al. [111] (2019) | Polyp Recognition | WCE | CNN | Private | Accuracy: 0.9319, precision: 74.51%, recall: 90.21%, F1: 81.83% |
| Nadimi et al. [114] (2020) | Colorectal polyps Localization and Autonomous Detection | WCE | CNN | Private | sensitivity: 98.1%, Accuracy: 98%, specificity: 96.3% |
| Javed et al. [70] (2020) | Detection of Cellular Community for Issue Phenotyping | Histopathology | Handcrafted, CNN | Public | Patch level separation: average F-score for CCT dataset: 94.5%, average F-score for CRC-TP dataset: 91% Patient level separation: average F-score for CRC-TP dataset: 84% |