Zitat
Abstract
Determining the mutation status of proto-oncogene B-Rapidly Accelerated Fibrosarcoma (BRAF) is crucial in melanoma for guiding targeted therapies and improving patient outcomes. While genetic testing has become more accessible, histopathological examination remains central to routine diagnostics, and an image-based strategy could further streamline the associated time and cost. In this study, we propose a new machine learning framework that integrates a large-scale, pretrained foundation model (Prov-GigaPath) with a gradient-boosting classifier (XGBoost) to predict BRAF-V600 mutation status directly from histopathological slides. Our approach was trained and cross-validated on the Skin Cutaneous Melanoma (SKCM) dataset from The Cancer Genome Atlas (TCGA; 275 slides), where the fine-tuned Prov-GigaPath model alone achieved an average Area Under the Curve (AUC) of 0.653 during cross-validation. An additional test on 68 slides from the University Hospital Essen (UHE), Germany, yielded an AUC of 0.697 (95 % CI: 0.553–0.821). Incorporating XGBoost significantly improved performance, reaching an AUC of 0.824 (SD=0.043) during cross-validation and 0.772 (95 % CI: 0.650–0.886) on the independent set—representing a new state-of-the-art for image-only BRAF mutation prediction in melanoma. By employing a weakly supervised, data-efficient pipeline, this method reduces the need for extensive annotations and costly molecular assays. While these results are not intended to replace genetic testing at this stage, they mark a new milestone in predicting BRAF mutation status solely from histopathological slides—a concept not yet fully established in prior research—and underscore the potential for seamlessly integrating automated, AI-driven decision-support tools into diagnostic workflows, thereby expediting personalized therapy decisions and advancing precision oncology.
Schlagwörter
Computational pathology
Deep learning
Foundation model
Machine learning
Melanoma
Whole slide image