Deep Learning Discovers New Morphological Features while Predicting Genetic Alterations from Histopathology of Papillary Thyroid Carcinoma

作者：Ingrid van Marion, Stefan Schulz, Christina Glasner, Jakob Nikolas Kather, Daniel Truhn, Markus Eckstein, Celine Mueller, Aurélie Fernandez, Simone Marquard, Marie Oliver-Metzig, Wilfried Roth, Matthias M. Gaida, Stephanie Strobl, Daniel‐Christoph Wagner, Arno Schad, Moritz Jesinghaus, Nils Hartmann, Thomas Johannes Musholt, Julia I. Staubitz, Sebastian Foersch · 发表于：Thyroid · 年份：2025 · DOI：10.1089/thy.2024.0691 · 被引用次数：9 · 研究领域：AI in cancer detection、Radiomics and Machine Learning in Medical Imaging、Biomedical Text Mining and Ontologies

Background:Papillary thyroid carcinoma (PTC) is the most common malignant tumor of the endocrine system. BRAF mutations occur in 40–60%, panRAS mutations in 10–15%, and different gene fusion events such as RET fusions in 7–35% of these neoplasms. Artificial intelligence (AI) methods could be used to predict genetic changes from conventional histopathological slides. Methods:In this retrospective study, we used two independent cohorts of patients with PTC, totaling 662 cases for the establishment of our AI pipeline. The Cancer Genome Atlas cohort (496 cases) served as the developmental cohort, while the Mainz cohort (166 cases) served as an independent external test cohort. BRAF, panRAS, and fusion status was determined for all of these patients as target variables. Vision Transformer was trained on digitized annotated hematoxylin and eosin-stained slides for the presence of these alterations. Highest probability image tiles were used to identify new morphological criteria associated with the genetic changes. Results:The trained model resulted in an area under the receiver operating characteristic curve of 0.882 (confidence interval 0.829–0.931) for BRAF, 0.876 (0.822–0.927) for panRAS, and 0.858 (0.801–0.912) for gene fusions. Accuracy was 79.3% (72.7–85.8%) for BRAF, 89.3% (84.2–94.0%) for panRAS, and 84.7% (78.8–90.2%) for gene fusions. The performance on the validation set was almost identical to that on the test set. Analyzing the highest predictive tiles, novel morpholog...