Scholarlink

Cancer remains one of the leading causes of mortality world-wide, necessitating accurate diagnosis and prognosis. Whole Slide Imaging (WSI) has become an integral part of clinical workflows with advancements in digital pathology. While Vision Transformers (ViT) have been applied to WSI analysis, their lack of interpretability limits clinical adoption. In this paper, we propose PATH-X, a deep learning framework that integrates pretrained ViT with SHAP (Shapley Additive Explanations) to enhance model explainability for patient stratification and risk prediction using WSIs from The Cancer Genome Atlas (TCGA). A representative image slice is selected for each WSI, and numerical feature embeddings are extracted using Google’s pre-trained ViT. These features are then compressed via an autoencoder and used for unsupervised clustering and classification tasks. Kaplan-Meier survival analysis evaluates risk stratification into two and three risk groups. SHAP is applied to identify key contributing features, which are mapped onto histopathological slices to provide spatial context. PATH-X was applied to three TCGA cancer types: kidney, glioma, and breast, selected for sufficient WSI sample sizes, and achieved robust stratification and strong classification performance across all cohorts, demonstrating the model’s effectiveness.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integrating Vision Transformers and Autoencoders for Interpretable Cancer Risk Assessment

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ahmad Hussein,
  • Ali Anaissi,
  • Mukesh Prasad,
  • Ali Braytee

摘要

Cancer remains one of the leading causes of mortality world-wide, necessitating accurate diagnosis and prognosis. Whole Slide Imaging (WSI) has become an integral part of clinical workflows with advancements in digital pathology. While Vision Transformers (ViT) have been applied to WSI analysis, their lack of interpretability limits clinical adoption. In this paper, we propose PATH-X, a deep learning framework that integrates pretrained ViT with SHAP (Shapley Additive Explanations) to enhance model explainability for patient stratification and risk prediction using WSIs from The Cancer Genome Atlas (TCGA). A representative image slice is selected for each WSI, and numerical feature embeddings are extracted using Google’s pre-trained ViT. These features are then compressed via an autoencoder and used for unsupervised clustering and classification tasks. Kaplan-Meier survival analysis evaluates risk stratification into two and three risk groups. SHAP is applied to identify key contributing features, which are mapped onto histopathological slices to provide spatial context. PATH-X was applied to three TCGA cancer types: kidney, glioma, and breast, selected for sufficient WSI sample sizes, and achieved robust stratification and strong classification performance across all cohorts, demonstrating the model’s effectiveness.