Table 2.
A summary of recent research applying AI to thyroid cytology specimens.
| Study | Year | Aim | Technique | Level | Sample Size | Reported Metrics | Results |
|---|---|---|---|---|---|---|---|
| Savala et al. (60) | 2018 | FTC vs FA | Neural network | Slide | 57 | Accuracy | 100% |
| Margari et al. (61) | 2018 | Predict TBS diagnosis | Classification and regression trees | Slide | 521 | Accuracy | 91% |
| Benign vs malignant | Accuracy Sensitivity Specificity |
93.0% 92.4% 93.6% |
|||||
| Sanyal et al. (62) | 2018 | PTC vs non-PTC | CNN | Image | 370 | Accuracy Sensitivity Specificity |
85.1% 90.5% 83.3% |
| Guan et al. (63) | 2019 | PTC vs benign | CNN | Slide | 279 | Accuracy | 95.0% |
| Image | 887 | Accuracy Sensitivity Specificity |
97.7% 100% 94.9% |
||||
| Maleki et al. (64) | 2019 | PTC vs NIFTPs and noninvasive EFV-PTC |
Support vector machine | Slide | 59 | Accuracy Sensitivity Specificity |
76.1% 72.6% 81.6% |
| Fragopoulos et al. (65) | 2020 | Benign vs malignant | Neural network | Slide | 447 | Accuracy Sensitivity Specificity |
95.1% 95.0% 95.1% |
| Elliott Range et al. (66) | 2020 | Benign vs malignant | Two CNNs | Slide | 908 | Sensitivity Specificity AUROC |
92.0% 90.5% 0.932 |
| Zhu et al. (67) | 2021 | Efficient follicular cell segmentation | CNN | Slide | 43 | Pixel Accuracy | 99.3% in 49.5 s |
| Image | 6,900 | Pixel Accuracy | 98.7% in 97.4 s | ||||
| Lin et al. (68) | 2021 | Fast segmentation of PTC | CNN | Slide | 131 | Accuracy Precision Recall |
99% 86% 94% |
| Dov et al. (69) | 2021 | Benign vs malignant | Two CNNs | Slide | 908 | AUROC Average Precision |
0.870 74.3% |
| Yao et al. (70) | 2022 | Benign vs FA | Gradient boosting and extra trees classifiers | Image | 800 | AUROC Accuracy Precision Recall |
0.75 71% 72% 71% |
| Dov et al. (71) | 2022 | Assess pathologist performance when using and not using a decision-support system | Screening software utilising two CNNs | Slide | 109 | Pairwise weighted kappa statistic | 0.924 |
FTC, follicular thyroid carcinoma; FA, follicular adenoma; TBS, The Bethesda System; PTC, papillary thyroid carcinoma; CNN, convolutional neural network; NIFTP, noninvasive follicular thyroid neoplasm with papillary-like nuclear features; EFV-PTC, encapsulated follicular variant of papillary thyroid cancer; AUROC, area under the receiver operating characteristic curve.
The level column describes whether metrics were calculated for full slides or extracted images.