Deep Learning Discovers New Morphological Features while Predicting Genetic Alterations from Histopathology of Papillary Thyroid Carcinoma
Authors' conclusion
Our study demonstrates that predicting genetic alterations in digitized histopathological slides using AI is feasible in patients with PTC. Our model showed high accuracy in predicting these changes, making it potentially suitable for pre-screening. Explainability approaches uncovered previously undescribed morphological patterns associated with certain genotypes.
Study context
- Population
- Patients with papillary thyroid carcinoma (PTC); TCGA developmental cohort (496 cases) and Mainz external test cohort (166 cases)
- Participants
- 662 patients2 cohorts
- Comparison
- BRAF vs panRAS vs Gene fusions
- Outcome
- Prediction of genetic alterations from H&E slides
Methodological detailsdeep learning (Vision Transformer), independent external test cohort
- Deep learning (Vision Transformer)
- Independent external test cohort
Performance
As reported in the abstract
| Test or model | AUC | Accuracy |
|---|---|---|
| BRAF | 0.882 (CI 0.829-0.931) | 79.3% (72.7-85.8%) |
| panRAS | 0.876 (0.822-0.927) | 89.3% (84.2-94.0%) |
| Gene fusions | 0.858 (0.801-0.912) | 84.7% (78.8-90.2%) |
AUC
BRAF0.882 (CI 0.829-0.931)
panRAS0.876 (0.822-0.927)
Gene fusions0.858 (0.801-0.912)
Accuracy
BRAF79.3% (72.7-85.8%)
panRAS89.3% (84.2-94.0%)
Gene fusions84.7% (78.8-90.2%)
Further findings
Morphological features
- Highest predictive tiles revealed novel morphological criteria for fusion-associated PTC
Original publication
Deep Learning Discovers New Morphological Features while Predicting Genetic Alterations from Histopathology of Papillary Thyroid Carcinoma. Thyroid. Published 3 July 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →