Wolfgang Gaggl PhD, Presenter: Researcher, Prism Clinical Imaging, Inc

Svyatoslav Vergun, Abstract Co-Author: Nothing to Disclose

Matthew Andreoli, Abstract Co-Author: Nothing to Disclose

Veena A. Nair PhD, Abstract Co-Author: Nothing to Disclose

Vivek Prabhakaran MD, PhD, Abstract Co-Author: Nothing to Disclose

Resting state functional MRI (rs-FMRI) enables clinicians to define critical areas and margins for pre-surgical planning of brain tumor resections without requiring the active participation of the patient. While task-based FMRI has gained utility in the clinical environment, rs-FMRI needs to be automatized and verified in tumor patients to be useful as a reliable clinical tool.

Data were acquired from 48 patients (24 with brain tumors, 24 epilepsy and vascular lesions) including rs-FMRI, task-based FMRI, diffusion tensor imaging (DTI) and structural MRI on 1.5T and 3T MRI scanners. Data were preprocessed (Allen EA, 2011) using AFNI (NIH, Bethesda, MD) and FSL (Oxford, UK) and decomposed into individual functional network components using independent component analysis (ICA) implemented in the GIFT toolbox (MRN, Albuquerque, NM) calculated for 28 and 75 components. ICA components were both manually identified by a trained radiologist overlaid on the anatomical and DTI images and compared by spatial correlation to published template components from healthy subjects (Calhoun, 2008). Predictive values from radiologist vs. automation where generated as well as ranked cross-correlation values.

Reproducible ICA components could be identified from both the 28 and 75 component analyses. Higher component numbers resulted in higher spatial detail and higher classifier values, but occasionally led to functional networks distributed across several components. The median classifier achieved better than 80% agreement. Using the non-deformable MNI registration to warp templates into subject space, templates showed considerable overlap with the tumor in some instances. Calculated ICA components, however, followed the outline of the tumor highlighting functional gray matter as classified by a clinician.

Our automated classification allows extraction of functional network components quickly with good agreement to the manual reader and with seamless integration into the existing clinical FMRI workflow. A larger functional component template library for use with clinical patient populations is currently underway for further validation and improvement of classification accuracy.

Task-free functional MRI can aid in identification of eloquent brain tissue in tumor resections by outlining functional networks and critical margins where active patient participation is not possible.

Gaggl, W, Vergun, S, Andreoli, M, Nair, V, Prabhakaran, V, Automated Task-Free Resting-State Functional MRI to Define Critical Margins in Surgical Planning for Brain Tumor Surgery.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14014615.html