Scholarlink

<p>Accurate tumor modeling is critical for personalized cancer treatment. This study proposes a hybrid framework combining deep learning, variational segmentation, and physics-informed modeling for precise tumor delineation and growth prediction. The pipeline integrates U-Net for semantic initialization with Chan–Vese active contours for boundary refinement, leveraging data-driven insights and geometric regularization. Tumor evolution uses a reaction–diffusion equation solved via physics-informed neural networks (PINNs) that embed governing equations into training. Bayesian optimization fine-tunes hyperparameters while Hamiltonian Monte Carlo infers parameter distributions. The framework quantifies epistemic and aleatory uncertainties for robust predictions. Fascinating results show high segmentation accuracy and strong agreement between predicted and observed growth patterns, validating the model as a digital twin for tumor analysis.</p>

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

Tumor Propagation Modeling Using Physics-Informed Neural Networks and Variational Methods for MRI Image Segmentation

  • 期刊论文

  • 发表日期：

    2026年2月15日

查看全文
  • Walid El Afari,
  • Mohammed Ziani,
  • Fatna Assaoui

摘要

Accurate tumor modeling is critical for personalized cancer treatment. This study proposes a hybrid framework combining deep learning, variational segmentation, and physics-informed modeling for precise tumor delineation and growth prediction. The pipeline integrates U-Net for semantic initialization with Chan–Vese active contours for boundary refinement, leveraging data-driven insights and geometric regularization. Tumor evolution uses a reaction–diffusion equation solved via physics-informed neural networks (PINNs) that embed governing equations into training. Bayesian optimization fine-tunes hyperparameters while Hamiltonian Monte Carlo infers parameter distributions. The framework quantifies epistemic and aleatory uncertainties for robust predictions. Fascinating results show high segmentation accuracy and strong agreement between predicted and observed growth patterns, validating the model as a digital twin for tumor analysis.