Marion Ingrid, Schulz Stefan, Glasner Christina, Kather Jakob Nikolas, Truhn Daniel, Eckstein Markus, Mueller Celine, Fernandez Aurélie, Marquard Simone, Oliver Metzig Marie, Roth Wilfried, Gaida Matthias Martin, Strobl Stephanie, Wagner Daniel-Christoph, Schad Arno, Jesinghaus Moritz, Hartmann Nils, Musholt Thomas Johannes, Staubitz-Vernazza Julia I, Foersch Sebastian
Institute of Pathology, University Medical Center Mainz, Mainz, Germany.
Medical Faculty Carl Gustav Carus, Else Kroener Fresenius Center for Digital Health, Technical University Dresden, Dresden, Germany.
Thyroid. 2025 Jul;35(7):771-780. doi: 10.1089/thy.2024.0691. Epub 2025 Jul 3.
Papillary thyroid carcinoma (PTC) is the most common malignant tumor of the endocrine system. mutations occur in 40-60%, mutations in 10-15%, and different gene fusion events such as fusions in 7-35% of these neoplasms. Artificial intelligence (AI) methods could be used to predict genetic changes from conventional histopathological slides. 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. , , 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. The trained model resulted in an area under the receiver operating characteristic curve of 0.882 (confidence interval 0.829-0.931) for , 0.876 (0.822-0.927) for , and 0.858 (0.801-0.912) for gene fusions. Accuracy was 79.3% (72.7-85.8%) for , 89.3% (84.2-94.0%) for , 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 morphological criteria for fusion-associated PTC could be discovered. 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. Providing pathologists with these AI-based features could improve their accuracy. Assuming further positive prospective validation, this discovery could contribute to a deeper understanding of PTC.
甲状腺乳头状癌（PTC）是内分泌系统最常见的恶性肿瘤。40%-60%的病例存在 突变，10%-15%的病例存在 突变，7%-35%的此类肿瘤存在不同的基因融合事件，如 融合。人工智能（AI）方法可用于从传统组织病理学切片预测基因变化。在这项回顾性研究中，我们使用了两个独立的PTC患者队列，共662例用于建立我们的AI流程。癌症基因组图谱队列（496例）作为开发队列，而美因茨队列（166例）作为独立的外部测试队列。对所有这些患者确定 、 和融合状态作为目标变量。针对这些改变的存在，在数字化注释的苏木精和伊红染色切片上训练视觉Transformer。使用概率最高的图像切片来识别与基因变化相关的新形态学标准。训练后的模型对于 的受试者工作特征曲线下面积为0.882（置信区间0.829-0.931），对于 为0.876（0.822-0.927），对于基因融合为0.858（0.801-0.912）。 的准确率为79.3%（72.7-85.8%）， 的准确率为89.3%（84.2-94.0%），基因融合的准确率为84.7%（78.8-90.2%）。验证集上的表现与测试集几乎相同。通过分析预测性最高的切片，可以发现融合相关PTC的新形态学标准。我们的研究表明，使用AI预测数字化组织病理学切片中的基因改变在PTC患者中是可行的。我们的模型在预测这些变化方面显示出高准确率，使其有可能适用于预筛查。可解释性方法揭示了与某些基因型相关的先前未描述的形态学模式。为病理学家提供这些基于AI的特征可以提高他们的准确率。假设进一步的前瞻性验证呈阳性，这一发现可能有助于更深入地了解PTC。
