Table 1.
Histopathologically related machine learning models used for clinical applications in GI cancers. Machine learning algorithms and models are grouped according to their specific computational task and GI cancer type to compare their performance and clinical application. The sources of the datasets and sample sizes are also summarized.
| Author | Task | Cancer Type | Type of WSI | Dataset | Algorithm/ Model |
Performance | Clinical Application |
|---|---|---|---|---|---|---|---|
| Yoshida et al. [24] | Classification | Gastric cancer | H&E | Training and testing: 3062 WSIs | e-Pathologist | Positive for carcinoma or suspicion of carcinoma vs. caution for adenoma or suspicion of a neoplastic lesion vs. negative for a neoplastic lesion Overall concordance rate: 55.6% Kappa coefficient: 0.28 (95% CI: 0.26–0.30) Negative vs. non-negative Sensitivity: 89.5% (95% CI: 87.5–91.4%) Specificity: 50.7% (95% CI: 48.5–52.9%) Positive predictive value: 47.7% (95% CI: 45.4–49.9%) Negative predictive value: 90.6% (95% CI, 88.8–92.2%) |
Differentiation and diagnosis gastric cancer grade |
| Yasuda et al. [25] | Classification | Gastric cancer | H&E | Training and testing: 66 WSIs | wndchrm | Noncancer vs. well-differentiated gastric cancer AUC: 0.99 Noncancer vs. moderately differentiated gastric cancer AUC: 0.98 Noncancer vs. poorly differentiated gastric cancer AUC: 0.99 |
Differentiation and diagnosis gastric cancer grade |
| Jiang et al. [106] | Classification and prognosis | Gastric cancer | H&E | Training: 251 patients Internal validation: 248 patients External validation: 287 patients |
Support vector machine | Patients might benefit more from postoperative adjuvant chemotherapy vs. patient might not postoperative adjuvant chemotherapy training cohort: 5-year overall survival AUC: 0.796 5-year disease-free survival AUC: 0.805 Internal validation cohort: 5-year overall survival AUC: 0.809 5-year disease-free survival AUC: 0.813 External validation cohort: 5-year overall survival AUC: 0.834 5-year disease-free survival AUC: 0.828 |
Prognosis of gastric cancer patients and identification of patients who might benefit from adjuvant chemotherapy |
| Cosatto et al. [26] | Detection | Gastric cancer | H&E | Training set: 8558 patients Test set: 4168 patients |
Semi-supervised multi-instance learning framework | Positive vs. negative AUC: 0.96 |
Detection of gastric cancer |
| Jiang et al. [27] | Classification | Colon cancer | H&E | Training: 101 patients Internal validation: 67 patients External validation: 47 patients |
InceptionResNetV2 + gradient-boosting decision tree machine classifier | High-risk recurrence vs. low-risk recurrence Internal validation hazard ratio: 8.9766 (95% CI: 2.824–28.528) External validation hazard ratio: 10.273 (95% CI: 2.177–48.472) Poor vs. good prognosis groups: Internal validation hazard ratio: 10.687 (95% CI: 2.908–39.272) External validation hazard ratio: 5.033 (95% CI: 1.792–14.132) |
Prognosis of stage III colon cancer |
WSI = whole-slide imaging; H&E = haematoxylin and eosin; CI= confidence interval; AUC = area under the curve; wndchrm = weighted neighbour distance using compound hierarchy of algorithms representing morphology.