报告题目：Illuminating Accuracy with Learning-guided Refinement: A Warm-basis Iterative Method for Fluorescence Molecular Tomography

报告人：姜嘉骅 伯明翰大学

报告时间：8月4日 10:00-11:00

报告地点：新楼308

摘要：

Fluorescence Molecular Tomography (FMT) is a non-invasive optical imaging technology widely used in biomedical research to reconstruct the volumetric distribution of fluorescent targets by measuring fluorescence signals emitted from the surface. FMT reconstruction faces significant accuracy challenges due to limited light penetration and strong photon scattering, which degrade depth resolution and worsen the inverse problem’s ill-posedness. Iterative methods struggle with poor Z-resolution despite advanced regularization, while operator-learning approaches can improve depth recovery but rely on large, high-quality paired datasets that are often impractical to acquire experimentally. Moreover, directly using the prediction from a trained deep learning model as an initial guess does not always lead to improved reconstructions. We present a warm-basis iterative projection method (WB-IPM) for solving 3D FMT problems, where the initial basis is obtained by a trained neural network, and then the reconstruction is refined by an augmented flexible hybrid projection method. Theoretical results show that, under reasonable conditions, our method can achieve a lower error bound than standard flexible hybrid projection methods. We demonstrate the effectiveness of our method on both numerical and real experiments.

报告人简介：

姜嘉骅，英国伯明翰大学助理教授，主要从事模型降阶，不确定性量化，反问题在图像处理上的应用等方面的研究。在医疗成像方面，姜博士提出的混合投影成像技术和基于深度学习的直接采样法在荧光成像，扩散光层析成像，CT，核磁共振成像上有重要应用。姜博士获中国科学技术大学学士学位，麻省大学达特茅斯分校博士学位，之后赴弗吉尼亚理工大学展开博士后研究。在SISC，JSC，Inverse problems 等多个应用数学和工程领域的核心期刊上发表论文，同时还担任JSC，JCP，Inverse problems等多个国际重要学术期刊的审稿人。