Dynamic PET Image Reconstruction via Self-Supervised Neural Network
数学专题报告
报告题目(Title)：Dynamic PET Image Reconstruction via Self-Supervised Neural Network
报告人(Speaker)：丁乔乔（上海交通大学）
地点(Place)：后主楼1124
时间(Time)：2025年11月9日（周日）16:00-17:00
邀请人(Inviter)：段玉萍
报告摘要
Dynamic PET imaging provides more comprehensive physiological information compared to conventional static PET imaging. Deep neural networks (DNNs) have emerged as a prominent tool in advancing medical image reconstruction methods. In this talk, I will present two works on dynamic PET/PET parametric imaging. In the first work, we proposed an unsupervised learning approach，Non-negative Implicit Neural Representation Factorization (NINRF). This method is based on low-rank matrix factorization of unknown images, where neural networks are employed to represent both coefficients and bases. In the second work, we developed a modified Logan reference plot model to reduce the acquisition time in dynamic PET imaging for parametric reconstruction. Extensive experiments on both simulated and clinical data demonstrate the effectiveness of our proposed methods.
主讲人简介
丁乔乔，上海交通大学自然科学研究院任助理研究员。主要研究方向是医学图像反问题的建模与医学图像分析，发表了多篇原创性高水平论文，发表的杂志有计算与应用数学杂志以及医工交叉学科杂志，包括 Inverse problems，SIIMS，IEEE TMI、 IEEE TCI, MICCAI, ICLR等，其中。一篇ICLR 工作 “基于标准化流的自编码器医疗图像异常检测算法” 入选 “2023 年度中国医学人工智能代表性算法” 三项算法之一。入选 2023 年上海市海外高层次人才计划，主持国家自然科学基金青年项目，参与国家重点研发计划青年科学家项目。
