A machine learning algorithm using diffusion tensor data for automated mapping of Wallerian degeneration in chronic stroke

Thursday, October 11, 2012 — Poster Session III
10:00 a.m. – Noon	Natcher Conference Center, Building 45	NICHD	IMAG-20

Authors

• P. Modi
• M. Irfanoglu
• E. Buch
• V. Buch
• M. Tobita
• C. Pierpaoli

Abstract

Diffusion Tensor Imaging (DTI) is used to investigate non-invasively microstructural features of brain tissue and their changes in disease. The anatomical abnormalities associated with chronic stroke and Wallerian degeneration are well known and well characterized with DTI. In this study, we aim at developing a machine learning algorithm (MLA) that could work with DTI data to map Wallerian degeneration in stroke patients and classify subjects into different pathology groups in a completely automated fashion. We trained our MLA with 28 patient and 20 healthy control datasets and performed cross-validation using the leave-one-out method. As DTI-derived metrics we used FA, Trace(D), and skewness measured in the cerebral peduncle, a region where motor pathways converge. As an encouraging preliminary result, we achieved 100% sensitivity and specificity in separating patients from controls. Our next goal is to train our MLA to identify the DTI-derived metrics that are most significant for classifying the patient population into subgroups with strokes from different vascular territories. If these encouraging preliminary findings are confirmed in more complex clinical situations, this method can become an important diagnostic tool for automated and unbiased mapping of regional abnormalities in a number of neurological and psychiatric disorders.

back to top