Biostatistics and Bioinformatics Monthly Seminar with Lin Zhang, PhD
Statistical image partitioning methods for lesion-wise cancer detection of prostate MRI
Lin Zhang, PhD
Associate Professor
Division of Biostatistics and Health Data Science
University of Minnesota
Meeting ID: 879 0675 3577
Passcode: 8Vbm9CCu
Abstract:
Imaging plays an important role in cancer diagnosis and staging by noninvasively evaluating the presence and extent of local and distant disease. Computer aided detection algorithms are being developed for fast and reproducible cancer diagnosis from complex and high-dimensional medical imaging data. While extensive statistical methods have been developed for voxel-wise cancer classification, existing lesion segmentation methods primarily rely on deep learning methods centered at the convolutional neural networks. We have developed novel statistical image partitioning methods for automatic lesion-wise cancer detection using imaging data, which jointly estimate the lesion boundaries and the spatial processes within each partitioned region in a Bayesian framework. We show through simulations and application to prostate cancer imaging data that the proposed methods well estimate the number and boundaries of cancerous regions of arbitrary number and shape with higher sensitivity and specificity compared to competitive methods.
This group meets monthly from Noon – 1 p.m. The meeting agenda includes a presentation, programmatic announcements and updates from Cancer Center program leaders.