勷勤数学•专家报告

题      目：RW-SPM: RegionWise Statistical Parametric Mapping

报  告  人：潘文亮 副研究员  (邀请人：葛文秀)

                                             中国科学院数学与系统科学研究院

时      间：6月30日  10：30-11：30

地     点：数科院西楼二楼会议室

报告人简介:

       现任中国科学院数学与系统科学研究院副研究员及博士生导师，专注于统计学习算法、医学图像数据分析和度量空间的非参数方法等领域研究。在Annals of Statistics、Journal of the American Statistical Association、ICML等统计学杂志和人工智能会议上发表了20篇以上学术论文，获得2022年教育部高等学校科学研究优秀成果自然科学类二等奖（排名第二）。主持的科研项目涵盖国家自然科学基金委青年基金(B类)、面上项目等。同时，担任北京生物医学统计与数据管理研究会副理事长，以及中国现场统计研究会统计交叉科学研究分会副秘书长。

摘      要：

      Statistical Parametric Mapping (SPM) is central to neuroimaging analysis, detecting associations between brain regions and clinical or genetic predictors. Yet anatomical variability, signal‐pattern heterogeneity, imperfect warping, and acquisition artifacts introduce misalignment errors that undermine accuracy and statistical power. To address these challenges, we propose Regionwise SPM (RW-SPM)—a novel framework that combines region‐based aggregation with adaptive smoothing guided by anatomical landmarks. By partitioning images into small, coherent regions and embedding their aggregated voxel distributions into a Hilbert space, RWSPM enables robust, distribution‐based association testing while remaining computationally efficient. Theoretically and in simulations, RWSPM markedly reduces misalignment bias, controls Type I error rates, and boosts sensitivity. Application to real neuroimaging data confirms its ability to improve both the accuracy and reliability of SPM results.