Improved Diagnostic Accuracy of Thyroid Fine-Needle Aspiration Cytology with Artificial Intelligence Technology
Authors' conclusion
These results suggest that the application of AI technology to thyroid FNA cytology may improve the diagnostic accuracy as well as intra- and inter-observer variability among pathologists.
Study context
- Population
- Thyroid FNA whole slide images (WSIs) from the Open artificial intelligence (AI) Dataset Project (3 university hospitals and 215 Korean institutions); 306 FNAs (78 malignant [PTC], 228 benign) from 86 institutions used for training10,332 image patches
- Participants
- 306 participants
- Comparison
- AI model vs expert cytopathologists for thyroid FNA cytology
- Outcome
- Diagnostic performance, AI vs cytopathologists (1031 randomly selected image patches)
Methodological detailsmulticenter, retrospective
- Multicenter
- Retrospective
Performance
As reported in the abstract
| Test or model | Sensitivity | Specificity | Accuracy |
|---|---|---|---|
| Inception ResNet v2 | 99.81% (95% CI 99.54-100.00%)Sensitivity on a scale from 0 to 100 percent | 99.61% (95% CI 99.23-99.99%)Specificity on a scale from 0 to 100 percent | 99.71% (95% CI 99.38-100.00%) |
| Expert cytopathologists | 87.26% (95% CI 85.22-89.30%)Sensitivity on a scale from 0 to 100 percent | 90.58% (95% CI 88.80-92.36%)Specificity on a scale from 0 to 100 percent | 88.91% (95% CI 86.99-90.83%) |
Sensitivity
Inception ResNet v299.81% (95% CI 99.54-100.00%)
Expert cytopathologists87.26% (95% CI 85.22-89.30%)
Sensitivity on a scale from 0 to 100 percentSpecificity
Inception ResNet v299.61% (95% CI 99.23-99.99%)
Expert cytopathologists90.58% (95% CI 88.80-92.36%)
Specificity on a scale from 0 to 100 percentAccuracy
Inception ResNet v299.71% (95% CI 99.38-100.00%)
Expert cytopathologists88.91% (95% CI 86.99-90.83%)
Further findings
| Outcome | Inception ResNet v2 |
|---|---|
| AI model performance in training, validation and test datasets | |
| Accuracy (training, validation, test) | 99.7%, 97.7%, 94.9% |
| Sensitivity (training, validation, test) | 99.9%, 99.6%, 100% |
| Specificity (training, validation, test) | 99.6%, 96.4%, 90.4% |
- AI model performance in training, validation and test datasets: best of 6 pretested AI models
Cytopathologists' performance after reference to AI results
- Accuracy of all experts after referring to AI
- 96%, 95%, and 96%
- Diagnostic agreement increased
- from 0.64 to 0.84
Original publication
Improved Diagnostic Accuracy of Thyroid Fine-Needle Aspiration Cytology with Artificial Intelligence Technology. Thyroid. Published 24 June 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →