Division of Sleep Medicine @ Harvard Medical School
Trainee Profile
Roman Torgovitsky
Graduate Student, Harvard School of Public Health
Research Unit(s)
Analytic and Modeling Unit, Division of Sleep Medicine, Brigham and Women's Hospital
Research Interests
I work on implementing and improving currently available multiple testing procedures and permutation methods for Sleep research.
Recently, sleep researchers became interested in using routinely collected EEG recordings to estimate intracerebral current densities. For example, studies have attempted visualization of alpha-delta sleep generators in different sleep stages, and revealed two distinct sleep spindle generators in the cortex. Another study investigated the cortical regions activated during REM sleep. These studies used the current statistical methodology to solve the source localization problem. After the cortex has been divided into thousands of voxels and current density has been estimated in each voxel, researchers are faced with a problem of testing thousands of hypotheses. For example, to determine which voxels in the cortex have been activated during REM sleep, scientists routinely obtain p-value for each voxel. Each p-value reflects how likely the corresponding voxel’s activation is different form zero (i.e., non-activated). Declaring a voxel significant based only on its p-value is not satisfactory, as this procedure inflates the number of false discoveries, and thus, doesn’t allow optimal localization of the signal.
Recently, sleep researchers became interested in using routinely collected EEG recordings to estimate intracerebral current densities. For example, studies have attempted visualization of alpha-delta sleep generators in different sleep stages, and revealed two distinct sleep spindle generators in the cortex. Another study investigated the cortical regions activated during REM sleep. These studies used the current statistical methodology to solve the source localization problem. After the cortex has been divided into thousands of voxels and current density has been estimated in each voxel, researchers are faced with a problem of testing thousands of hypotheses. For example, to determine which voxels in the cortex have been activated during REM sleep, scientists routinely obtain p-value for each voxel. Each p-value reflects how likely the corresponding voxel’s activation is different form zero (i.e., non-activated). Declaring a voxel significant based only on its p-value is not satisfactory, as this procedure inflates the number of false discoveries, and thus, doesn’t allow optimal localization of the signal.
Mentor(s)
Victor DeGruttolla PhD, Elizabeth B. Klerman, MD, PhD
Teaching
Teaching assistant: Introduction to Biostatistics, Statistical Inference
Site Map | Contact Us | © 2006 by the President and Fellows of Harvard College
