【专家简介】：刘熔洁，现任佛罗里达州立大学统计系助理教授，于2010年在东南大学数学系获得信息与计算科学学士学位，2016年在美国德州大学圣安东尼奥分校获得电子与计算机工程博士学位，2020年在美国莱斯大学获得统计学博士学位。研究方向包括：贝叶斯统计，计算机视觉，机器/深度学习，强化学习等。在高水平国际统计神经科学期刊以及图像模式识别顶会上（Neurocomputing, NeuroImage, Frontiers in Neuroscience, IEEE Transactions on Neural Networks and Learning Systems, CVPR）发表学术论文40多篇。
【报告摘要】：To better understand and treat Alzheimer's Disease (AD), many types of research have been conducted to detect abnormal brain regions that can facilitate providing targeted medicine and improve the treatment pathways. However, these regions may vary among subjects due to heterogeneity arising from demographic factors such as age and gender. Furthermore, brain cells within a subject have inherent spatial dependence among themselves and a diseased cell may affect its neighboring cells to an unknown extent. In addition, unmeasured confounders and measurement errors can partially or completely mask the abnormal regions. All these points make these diseased regions very difficult to detect. To this end, we propose a Patient-specific Abnormal Region Detection (PARD) algorithm to identify the heterogeneous diseased regions by solving a Bayesian latent-space variable selection problem. Using Bayesian hierarchical modeling, we account for the heterogeneity among the subjects as a large-scale variability and incorporate the inherent spatial dependence within subjects using spike-and-slab priors into the latent space. A Gibbs sampling framework is derived for estimating the model parameters and hyper-parameters efficiently. The simulation study shows the superiority of the proposed algorithm over popular unsupervised learning methods. The algorithm is further applied to the resting-state MRI brain scans of subjects collected from Alzheimer's Disease Neuroimaging Initiative (ADNI) and the detected regions are validated by cross-matching with the brain's default mode network (DMN).
时 间：2023年6月13日 14:30
地 点：位育楼417
