Abstracts

Rationale: Numerous techniques exist for automatic segmentation of brain subcortical regions from MRI data, the majority of which necessitate the use of three-dimensional (3D) MRI scans. MRIs obtained in clinical trials in non-specialist hospitals are typically two-dimensional (2D). The use of automated volumetric techniques for 2D scans has been challenging and potentially leads to a lost opportunity to perform quantitative image analysis in context clinical trial studies. This investigation sought to modify a commonly used subcortical segmentation technique developed for use with 3D MRI scans for application to 2D scans in patients with epilepsy. The consistency in volume estimation between 2D and 3D scans in the same patients were analyzed. Methods: 2D (voxel size 0.4mmx0.4mmx3mm) and 3D (1mmx1mmx1mm) T1-weighted MRI scans were acquired in 24 patients with idiopathic generalized epilepsy. Image segmentation and volume estimation was performed on 2D and 3D scans using FIRST (FMRIB’S Integrated Registration and Segmentation Tool). For successful segmentation of the 2D scans, images were re-oriented to standard orientation, cropped and registered to an edited template using FLIRT and interpolated to 1mmx1mmx1mm resolution. Additionally, the Montreal Neurological Institute (MNI) standard template brain used in processing was cropped for consistency. FIRST was used to automatically identify the caudate nucleus, globus pallidus, putamen, hippocampus, and thalamus in both cerebral hemispheres. 2D segmentation reliability was assessed through comparison with 3D segmentations using intraclass correlation coefficients (ICCs) and through Dice coefficients. Results: Based on previous reports, ICCs were deemed acceptable if above 0.75 (Koo and Li, 2016, J Chiropr Med. 5(2):155-63). ICCs were above 0.75 for all regions of interest: left hemisphere caudate = 0.772, globus pallidus = 0.831, hippocampus = 0.798, putamen = 0.878, thalamus = 0.766, right hemisphere caudate = 0.825, globus pallidus = 0.756, hippocampus = 0.810, putamen = 0.756 and thalamus = 0.815 (mean = 0.808). Mean Dice coefficients for each region of interest were all above 0.7: left hemisphere caudate = 0.762, globus pallidus = 0.754, hippocampus = 0.707, putamen = 0.827, thalamus = 0.856, right hemisphere caudate = 0.752, globus pallidus = 0.766, hippocampus = 0.7, putamen = 0.811, and thalamus = 0.869 (mean = 0.78). Conclusions: Our results indicate that adapted pipelines using FIRST segmentation can be applied to the majority of 2D scans. We will endeavour to apply this modified pipeline to 2D MRI data acquired in context of the SANAD II clinical trial (Marson, A. et al., 2013) to investigate biomarkers of treatment outcome in newly diagnosed epilepsy. Funding: Funding for this project is sourced from the University of Liverpool and the Walton Centre NHS Foundation Trust.