Presentation

Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks

Using an automated segmentation of the language pathways for neurosurgical planning

Poster D42 in Poster Session D, Wednesday, October 25, 4:45 - 6:30 pm CEST, Espace Vieux-Port

Rachel Eliyahu¹, Donna Abecasis², Vered Kronfeld-Duenias^1,2, Benjamin Menashe¹, Irit Shapira-Lichter^2,3, Michal Ben-Shachar¹; ¹Bar-Ilan University, ²Functional MRI Center, Beilinson Hospital, ³Tel Aviv University

Presurgical brain imaging provides neurosurgeons with tools to better plan their access strategy to a brain tumor (Potgieser et al., 2014). To date, diffusion MRI data is often measured pre-surgically, followed by deterministic tractography and manual segmentation of white matter tracts (Wakana et al., 2007). Replacing manual segmentation with automatic segmentation and quantification tools is desirable because it is faster, cheaper, and more objective. This would allow expanding these methods to other patient populations. Automatic procedures for segmentation and quantification of white matter tracts have been applied to data from neurological populations (Deng et al., 2021). However, to our knowledge, such methods have not been systematically compared with manual segmentation procedures in patients with brain lesions. Validating automatic segmentations against manual methods is a critical step before adopting the automatic tools as part of the clinical protocol. In this study, we quantitatively compared automatic and manual tract segmentations in 31 patients with temporal, frontal or parietal lesions (mean age: 43y ±14, 18 males, 26 with left hemisphere lesions). Diffusion MRI data were collected in a 3T Philips scanner, using a diffusion-weighted, single-shot EPI sequence (32 diffusion directions at b = 1000 and 1 volume at b = 0 s/mm^2, voxel size: ~2*2*2mm^3). For each patient we identified four language-related tracts and their right hemisphere homologs: the arcuate fasciculus, inferior fronto-occipital fasciculus, inferior longitudinal fasciculus and uncinate fasciculus. For automatic segmentation, we used the Automatic Fiber Quantification (AFQ) package (Yeatman et al., 2012). Manual segmentations followed a published protocol (Wakana et al., 2007). Both methods showed a similar identification ratio: The automatic method identified 229/248 tracts, the manual method identified 225/248 tracts. Visual inspection of individual tracts showed a good fit between the methods generally, and, in some cases, higher sensitivity of the automatic methods. Quantitative estimates (fractional anisotropy and mean diffusivity) derived from automatic and manual segmentations of left hemisphere tracts were very highly correlated (r = 0.92-0.99). Some right hemisphere tracts showed slightly lower correlation values (r = 0.77-0.99). Anisotropy profiles of manually and automatically identified tracts overlapped closely in most of the left hemispheric tracts, while the right hemispheric tracts and the left uncinate fasciculus showed significant differences between the methods. In sum, the results so far provide encouraging data that support using automatic segmentation methods in patients with brain lesions, followed by visual inspection and minimal manual editing. In most patients, the lesions were located in the left hemisphere. Therefore, differences found in segmentations of the right hemispheric tracts are of lesser significance for presurgical planning. Future studies will examine the effects of lesion location, size, type and grade on the fit between tract segmentation methods.

Topic Areas: Disorders: Acquired,