2020-12-24
12月22日上午，美国芝加哥大学潘晓川教授作客国家数学与交叉科学中心十周年论坛，带来综合论坛系列的第九场线上报告，题目为“Non-Convex Optimization Problem in Image Reconstruction in Quantitative Dual-Energy Computed Tomography”。报告由中科院数学院常务副院长、交叉中心副主任高小山研究员主持。
当前成像技术在医学诊断领域得到了广泛的应用，因此发展先进的医学成像方法对精准的医学诊断具有非常重要的意义。在本次报告中，潘教授首先介绍了X射线计算机断层成像（CT）的硬件结构和单能CT图像重构的数学原理，描述了从扫描数据到图像重构，再到图像处理和分析，最终到应用的工作流程，进而阐明了其中学术研究与工业研发的区别和联系。随后，潘教授着重阐释了双能CT的物理原理、实际意义，以及图像重构算法。事实上，X射线光源发射的光束具有能谱效应，而同一物质对光子的线性衰减系数的大小依赖于光子的能量。考虑X射线的能谱效应，结合拟重构图像的先验信息，潘教授所构造的图像重构的数学模型是一个非凸优化模型，这等价于求解一个非线性反问题。为了求解此问题，潘教授提出了一个非凸primal-dual算法，并利用大量的模拟和实际的数据验证了该算法具有数值收敛性，同时指出该方法即使在采样数据减半的情况下也能获得理想的重构效果。最后，潘教授列出了目前国际上多个研究组采用机器学习/神经网络/深度学习/人工智能等方法得到的CT图像重构结果，同时表示到目前为止还没有证据能够证明这些方法可以求解图像重构反问题。活动结束前，报告人对观众提问进行了详细解答。
Xiaochuan Pan is a Professor in the Department of Radiology, Department of Radiation & Cellular Oncology, the Committee on Medical Physics, the Comprehensive Cancer Center, and the College at The University of Chicago. His research centers on physics, algorithms, and engineering underpinning tomographic imaging and its biomedical and clinical applications. He has developed close clinical and industrial collaboration and established robust translational programs. He is the recipient of awards such as IEEE NPSS Early Achievement Award and IEEE EMBS Technical Award, a Fellow of AAPM, AIMBE, IAMBE, IEEE, OSA, and SPIE, and Distinguished Investigator Award of ARR. Dr. Pan has served as a chair and/or reviewer of study sections/review panels for funding agencies, including NIH, NSF, DOE, NSERC, NSFC, and ASF. He is the Editor-in-Chief of IEEE Trans. Biomed Eng., and has served or is serving as an associate editor (or editorial board member) for journals in the field such as IEEE Trans. Med. Imaging, IEEE Trans. Biomed Eng., IEEE J. Transl. Eng. Health and Med., Phys. Med. Biol., Med. Phys., and J. Med. Imaging, as a chair/member of technical committees of professional organizations such as IEEE, RSNA, and AAPM, and as a general chair, chair/member of programs, themes, and technical/scientific committees for conferences such as IEEE EMBC, IEEE MIC, RSNA, AAPM, and MICCAI.
