Table 4.
Performance data on studies for early detection of EC using a ML
| Author | Cancer Type | Modality | algorithm | image | patient | AUROC | accuracy | sensitivity | specificity |
|---|---|---|---|---|---|---|---|---|---|
| Lou et al. |
EAC & ESCC |
CT | U-Net | 80 | - | - | - | - | - |
| Ghatwary et al. |
EAC & ESCC |
WLI |
Faster R-CNN SSD |
100 | 39 | - | - | 96% | 92% |
| Tang et al. |
EAC & ESCC |
WLI | MTCS | 805 | 255 | - | 93.43% | 92.82% | 96.20% |
| Ghatwary et al. | EAC | - | Faster R-CNN | 1000 | - | - | - | - | - |
| Yu et al. |
EAC & ESCC |
endoscopy images | MTL | 1003 | - | - | 96.96% | 95.64% | 97.70% |
| Wu et al. |
EAC & ESCC |
WLI |
Faster-RCNN DSN |
1051 | - | - | 96.28% | 90,34% | 97,18% |
| Liu et al. |
EAC & ESCC |
WLI | CNN | 1272 | 748 | - | 85.83% | 94.23% | 94.67% |
| Groof et al. | EAC | WLI | hybrid ResNet-Unet | 1704 | 669 | - | 89% | 90% | 88% |
| Tang et al. | ESCC | - | DCNN | 4002 | 1078 | 95,4% | 91.30% | 97.9% | 88.6% |
| Meng et al. | ESCC | WLI | YOLO v5 | 4447 | 837 | 98,2% | 92.9% | 91.90% | 94.7% |
| Gong et al. |
EAC & ESCC |
WLI | No-code deep-learning tool “Neuro-T” version 2.3.2 | 5162 | - | 95% | 95.6% | - | - |
| Shiroma et al. | ESCC |
WLI & NBI |
SSD | 8428 | - | - | 98% | 100% | 100% |
| Du et al. |
EAC & ESCC |
- |
RWS ECA-DDCNN |
20,965 | 4,077 | 98.77% | 90.63% | - | - |
| Putten et al. | EAC | endoscopy images |
U-Net Transfer Learning |
494,356 | - | - | 87.50% | 92.50% | 82.50% |
| Gan et al. |
EAC & ESCC |
OCT image | D-UCN | - | - | - | 98% | - | - |
| Wang et al. |
EAC & ESCC |
endoscopy & ultrasound | Cascade RCNN | - | 80 | - | 83% | - | - |
| Sui et al. |
EAC & ESCC |
CT | V-Net | - | 414 | - | 65% | 88.80% | 90.90% |
| Takeuchi et al. |
EAC & ESCC |
CT | CNN- VGG16 | - | 457 | - | 84.20% | 71.70% | 90.00% |
| Ghatwary et al. |
EAC & ESCC |
video | 3DCNN | - | - | - | 91.10% | - | - |
| Alharbe et al. |
EAC & ESCC |
image | Deep transfer learning | - | - | - | 99.7% | 99.49% | 99.78% |
| Zhao et al. |
EAC & ESCC |
digestive endoscopy |
Google Net V3 TensorFlow 1.6 |
- | 300 | 91% | 91.00% | 90.00% | 92.0% |
| Collins et al. |
EAC & ESCC |
- | SVM, MLP, 3DCNN | - | 10 | 93% | - | - | - |
| Zhao et al. |
EAC & ESCC |
- | CNN | - | 500 | - | - | 98% | 99,6% |
| Chen et al. |
EAC & ESCC |
- | Faster RCNN | 1520 | 421 | - | 92.15% | - | - |
| Tsai et al. |
EAC & ESCC |
WLI & NBI |
SSD VGG-16 |
155 153 |
- | - | 86% | 92% | - |
| Tsai et al. |
EAC & ESCC |
WLI | SSD | 1780 | - | - | 96.1% | 81.6% | - |
| Sali et al. | EAC | whole-slide tissue histopathology images (WSIs) | ResNet34 | 387 | 130 | - | - | - | - |
| Wang et al. |
EAC & ESCC |
WLI & NBI |
SSD |
498 438 |
- | - | 90.90% | 96.20% | 70.40% |
| Zhang et al. |
EAC & ESCC |
- |
Faster R-CNN VGG16 |
6445 | 200 | - | 90.3% | 92.5% | 88.70% |
| Guo et al. | ESCC | NBI | SegNet | 6473 | - | - | - | 98.04% | 95.03% |
| Fang et al. |
EAC & ESCC |
WLI & NBI |
U-Net |
75 91 |
- | - | 84.72% | - | - |