关键词: 甲状腺癌, BRAF-V600E突变, 人工智能

Abstract: Objective To investigate the application value of a deep learning model based on whole-slide imaging （WSI） in predicting the BRAF-V600E mutation status in thyroid cancer and to analyze the correlation between this mutation and patients' clinical characteristics. Methods A deep learning model based on a pre-trained ResNet50 network was developed using data from The Cancer Genome Atlas （TCGA）. The model employed a self-attention mechanism to extract key features from WSI and constructed a single-task binary classification model to predict the BRAF-V600E mutation status. A total of 305 papillary thyroid carcinoma （PTC） cases were used for model training, and 131 cases were used for validation. The model's performance was evaluated using the area under the receiver operating characteristic curve （AUC）. Additionally, the association between BRAF-V600E mutation and clinical characteristics such as gender, age, and tumor staging were analyzed. Results The model achieved AUC values of 0.972 and 0.904 on the training and validation datasets, respectively, demonstrating high predictive accuracy. Clinical characteristic analysis revealed that the BRAF-V600E mutation was more common in females, patients under 55 years of age, and those with advanced-stage tumors. The mutation was significantly associated with higher tumor stages and lymph node metastasis. Conclusion The deep learning-based WSI model performed excellently in predicting BRAF-V600E mutation status in thyroid cancer, providing support for personalized diagnosis and treatment. Future studies should integrate multi-center datasets to further validate the model's clinical applicability and generalizability.

Key words: thyroid cancer, BRAF-V600E mutation, artificial intelligence