Table 3.
A summary of recent research applying AI to histology specimens.
| Study | Year | Aim | Technique | Level | Sample Size | Reported Metrics | Results |
|---|---|---|---|---|---|---|---|
| Jothi and Rajam (90) | 2017 | PTC vs normal thyroid | Ensemble learning with two support vector machines and a closest-matching-rule classifier | Image | 219 | Accuracy Sensitivity Specificity |
99.5% 100% 98.6% |
| Wang et al. (91) | 2019 | Multiclassification of thyroid nodules | CNN | Slide | 806 | Accuracy | 98.4% |
| Image | 11,715 | Accuracy | 97.3% | ||||
| Tsou and Wu (92) | 2019 | Predict BRAF or RAS mutational status | CNN | Slide | 103 | Accuracy | 95.2% |
| Image | 2,595 | AUROC | 0.951 | ||||
| Dolezal et al. (93) | 2020 | Classify NIFTP, PTC-EFG, and PTC as BRAF- or RAS-like | CNN | Slide | 612 | – | – |
| Predict BRAF-RAS score and use it to discriminate NIFTP status | Slide | 497 | AUROC | 0.99 | |||
| Liu et al. (94) | 2021 | PTC vs normal thyroid | CNN | Image | 2,772 | Accuracy | 98.6% |
| Esce et al. (95) | 2021 | Identify lymph nodal metastases | CNN | Slide | 174 | Sensitivity Specificity AUROC |
94% 100% 0.964 |
| El-Hossiny et al. (96) | 2021 | Multiclassification of thyroid nodules | Cascaded CNNs | Image | 18,653 | Accuracy | 94.7% |
| Han et al. (97) | 2021 | PTC vs normal thyroid | CNN | Image | 16,500 | Sensitivity Specificity |
95.8% 95.1% |
| Anand et al. (98) | 2021 | Predict BRAF mutational status | Weakly supervised CNN | Slide | 529 | AUROC | 0.98 |
| Böhland et al. (99) | 2021 | PTC-like (PTC, NIFTP and FV-PTC) vs non-PTC-like (FA, FTC) | CNNs and machine learning algorithms applied to two datasets | Slide | 156 | Accuracy | 89.7% |
| Slide | 133 | Accuracy | 83.5% | ||||
| Deng et al. (100) | 2022 | PTC vs non-PTC | Ensemble of a CNN and random forest | Image | 610 | Accuracy Sensitivity Specificity AUROC |
93.8% 85.9% 97.2% 0.982 |
| Stenman et al. (101) | 2022 | Quantification of tall cells in PTC | Two CNNs | Image | 2,970 | Sensitivity Specificity |
93.7% 94.5% |
PTC, papillary thyroid carcinoma; CNN, convolutional neural network; BRAF, RAS, gene types; AUROC, area under the receiver operating characteristic curve; NIFTP, noninvasive follicular thyroid neoplasm with papillary-like nuclear features; PTC-EFG, papillary thyroid carcinoma with extensive follicular growth; FV-PTC, follicular variant of papillary thyroid carcinoma.