Scholarlink

<p>Dosing of chemotherapy agents is a central challenge in oncology because treatment schedules must balance tumor suppression with toxicity risk. Conventional PK–PD models commonly evaluate treatment response using tumor-burden or volumetric trajectories, although structural tumor reorganization may occur before measurable volumetric reduction. In this study, we propose a topology-regularized physics-informed neural network framework, termed TR-PINN-Chemo, for computational chemotherapy dose optimization under hematological safety constraints. The framework combines a mechanistic PK–PD–toxicity model, neural differential equation learning, topology-derived descriptors from tumor masks, and DNA-inspired evolutionary dose search. In a synthetic cohort of virtual breast cancer patients, the optimized schedule produced greater normalized tumor-burden reduction than the standard schedule while maintaining the predefined neutrophil safety threshold. Topology-derived descriptors suggested earlier structural reorganization than volume-based response in the controlled simulation setting. A preliminary breast MRI feasibility assessment further demonstrated that Euler characteristic and Betti numbers can be extracted from segmented tumor masks, although this imaging component was not designed as a fully powered clinical validation study. These findings support the potential methodological value of integrating topology-derived structural information with PINN-based treatment modeling, while prospective clinical validation and more comprehensive toxicity modeling are required before clinical application.</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

Physics-informed neural networks for personalized chemotherapy optimization with topological tumor analysis

  • 期刊论文

  • 发表日期：

    2026年6月17日

查看全文
  • A. Y. Xani,
  • N. Yildirim,
  • Jordan Jean Faustin,
  • Cameron Paxton,
  • Azaria Johnson

摘要

Dosing of chemotherapy agents is a central challenge in oncology because treatment schedules must balance tumor suppression with toxicity risk. Conventional PK–PD models commonly evaluate treatment response using tumor-burden or volumetric trajectories, although structural tumor reorganization may occur before measurable volumetric reduction. In this study, we propose a topology-regularized physics-informed neural network framework, termed TR-PINN-Chemo, for computational chemotherapy dose optimization under hematological safety constraints. The framework combines a mechanistic PK–PD–toxicity model, neural differential equation learning, topology-derived descriptors from tumor masks, and DNA-inspired evolutionary dose search. In a synthetic cohort of virtual breast cancer patients, the optimized schedule produced greater normalized tumor-burden reduction than the standard schedule while maintaining the predefined neutrophil safety threshold. Topology-derived descriptors suggested earlier structural reorganization than volume-based response in the controlled simulation setting. A preliminary breast MRI feasibility assessment further demonstrated that Euler characteristic and Betti numbers can be extracted from segmented tumor masks, although this imaging component was not designed as a fully powered clinical validation study. These findings support the potential methodological value of integrating topology-derived structural information with PINN-based treatment modeling, while prospective clinical validation and more comprehensive toxicity modeling are required before clinical application.