Scholarlink

<p>Accurate characterization of brain tissue microstructure using diffusion MRI (dMRI) depends on data acquisition protocols. While optimal design has been explored through Fisher information, empirical subsampling of large datasets and machine learning strategies, a unified approach of optimization for diverse white and gray matter biophysical models remains necessary. To this end, we propose a generalized protocol design framework based on the Cramér-Rao Lower Bound (CRLB), which accommodates user-defined constraints while comprehensively optimizing across parameter spaces. By implementing biophysical models in a differentiable environment to compute gradients using automatic differentiation, we avoid the approximation errors of finite-difference methods and the scalability limitations of derivative-free optimization. This enables stable, reproducible convergence for high-dimensional, nonlinear models. Moreover, this approach produces data acquisition protocols that improve the estimation accuracy of microstructural metrics such as axonal diameter indices, diffusivities, and compartment water-exchange rates, thereby enhancing the fidelity of biophysical modeling for research and clinical applications.</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

Diffusion MRI experimental design optimization for microstructure imaging

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月19日

查看全文
  • Hamza Farooq,
  • Yongxin Chen,
  • Ghulam Rasool,
  • Tryphon Georgiou,
  • Christophe Lenglet

摘要

Accurate characterization of brain tissue microstructure using diffusion MRI (dMRI) depends on data acquisition protocols. While optimal design has been explored through Fisher information, empirical subsampling of large datasets and machine learning strategies, a unified approach of optimization for diverse white and gray matter biophysical models remains necessary. To this end, we propose a generalized protocol design framework based on the Cramér-Rao Lower Bound (CRLB), which accommodates user-defined constraints while comprehensively optimizing across parameter spaces. By implementing biophysical models in a differentiable environment to compute gradients using automatic differentiation, we avoid the approximation errors of finite-difference methods and the scalability limitations of derivative-free optimization. This enables stable, reproducible convergence for high-dimensional, nonlinear models. Moreover, this approach produces data acquisition protocols that improve the estimation accuracy of microstructural metrics such as axonal diameter indices, diffusivities, and compartment water-exchange rates, thereby enhancing the fidelity of biophysical modeling for research and clinical applications.