Bayesian image-based clinical prediction in the longitudinal setting

Speaker: Mert Sabuncu , Martinos Center for Biomedical Imaging, MGH, Harvard Medical School

Date: Monday, January 25, 2016

Time: 4:00 PM to 5:00 PM

Public: Yes

Location: 32-D507

Event Type:

Room Description:

Host: Polina Golland, CSAIL

Contact: Polina Golland,

Relevant URL:

Speaker URL: None

Speaker Photo:

Reminders to:

Reminder Subject: TALK: Bayesian image-based clinical prediction in the longitudinal setting

Longitudinal studies, where data are collected at multiple time
points, are becoming increasingly widespread. However, biomedical
applications of machine learning algorithms have largely neglected the
longitudinal design. Many of the tools are built on the assumption
that each datapoint is an independent sample or the prediction target
is a constant variable. In this talk, I will present some
observations on longitudinal neuroimaging studies and how these data
are often analyzed using classical statistical tools. I will then
move on to show some recent/on-going work where I have developed novel
machine learning algorithms that were inspired by these classical
methods. I will consider two applications: (1) prediction of the time
of a future event (e.g., symptom onset) based on a clinical image
scan, and (2) clinical prediction based on longitudinal image data. I
will be using the framework of the Relevance Voxel Machine (RVoxM),
which is a sparse Bayesian learning method we developed to compute
predictions from image data.

Research Areas:

Impact Areas:

See other events that are part of the Biomedical Imaging and Analysis 2015/2016.

Created by Polina Golland Email at Thursday, October 22, 2015 at 3:00 AM.