BISON: Brain tISue segmentatiON pipeline using T1-weighted magnetic resonance images and a random forests classifier

Abstract

Introduction: Accurate differentiation of brain tissue types from T1-weighted magnetic resonance images (MRIs) is a critical requirement in many neuroscience and clinical applications. Accurate automated tissue segmentation is challenging due to the variabilities in the tissue intensity profiles caused by differences in scanner models and acquisition protocols, in addition to the varying age of the subjects and potential presence of pathology.In this paper, we present BISON (Brain tISue segmentatiON), a new pipeline for tissue segmentation.Methods: BISON performs tissue segmentation using a random forests classifier and a set of intensity and location priors obtained based on T1-weighted images. The proposed method has been developed and cross-validated based on multi-center and multi-scanner manual labels of 72 subjects aging from 5-96 years old, ensuring the generalizability of the results to new data from various age ranges. In addition, we assessed the test-retest reliability of BISON on 2 datasets; a. using 20 subjects that had scan/re-scan MRIs and manual segmentations available, and b. using a human phantom dataset including 90 scans from a single individual acquired across 10 years. Results:The results of the proposed method were compared against Atropos, a commonly used tissue classification method from ANTs. The proposed method yielded cross-validation Dice Kappa values of κGM = 0.88 ± 0.03, κWM = 0.85 ± 0.03, κCSF = 0.77 ± 0.11, outperforming ANTs Atropos (κGM = 0.79 ± 0.05, κWM = 0.84 ± 0.05, κCSF = 0.64 ± 0.22) as well as test-retest Dice Kappa values of κGM = 0.94 ± 0.006, κWM = 0.92 ± 0.006, κCSF = 0.77 ± 0.11 outperforming both manual (κGM = 0.92 ± 0.01, κWM = 0.91 ± 0.01, κCSF = 0.74 ± 0.03) andANTs Atropos (κGM = 0.87 ± 0.001, κWM = 0.92 ± 0.001, κCSF = 0.79 ± 0.05). The human phantom dataset validations showed high generalizability for both Atropos (κGM = 0.97 ± 0.01, κWM = 0.96 ± 0.01, κCSF = 0.93 ± 0.02) and BISON (κGM = 0.95 ± 0.01, κWM = 0.94 ± 0.01, κCSF = 0.85 ± 0.03), while Atropos tended to consistently under-segment the cortical CSF. Finally, our assessment of BISON, Atropos, FAST from FSL, and SPM12 segmentations in presence of white matter hyperintensities (WMHs) showed that BISON outperforms the other three methods, correctly detecting WMHs as WM. Conclusion:Our results show that BISON can provide accurate and robust segmentations in data from different age ranges and various scanner models, making it ideal for performing tissue classification in large multi-center and multi-scanner databases.

Conversion of diffusely abnormal white matter to focal lesions is linked to progression in secondary progressive multiple sclerosis

“…All pipelines used in the processing of the images have been developed and validated for use in multi-center and multi-scanner datasets and have been previously used in many such applications. 19,18,24…”

Diffusely abnormal white matter converts to T2 lesion volume in the absence of acute inflammation

“…All scans were processed using our standard image processing pipeline for longitudinal data, extensively described in our previous work 10,12 . Briefly, the steps applied to the MRI data at each timepoint are: 1) Intensity non-uniformity correction 14 ; 2) linear intensity normalization; 3) brain tissue extraction 15 ; 4) tissue classification using T1w images 16 ; 5) rigid co-registration of T2w to T1w images; 6) linear and nonlinear registration of the T1w images to MNI-ICBM152-2009c template 17,18 ; 7) automatic segmentation of WM lesions using a Bayesian Classifier, used to mask lesion voxels and obtain the intensity mode of the NAWM voxels 19 ; 8) automatic separation of FWML from DAWM using intensity thresholds based on normalized T2 intensity mode values (the details of this automated separation techniques have been described in our previous studies) 10,12 .…”

Reliability Assessment of Tissue Classification Algorithms for Multi-Center and Multi-Scanner Data

“…Tissue segmentations were performed after these preprocessing steps using: 1) Atropos (Avants et al, 2011); 2) BISON (Dadar and Collins, 2019); 3) Classify_Clean (Cocosco et al, 2003); FAST 5.0 (Zhang et al, 2001); FreeSurfer 6.0.0 (Fischl, 2012); and SPM12 (Penny et al, 2011). For all pipelines, default settings were used.…”

“…Brain tISue segmentatiON (BISON) is an open source pipeline based on a random forests classifier that has been trained using a set of intensity and location features from a multi-center manually labelled dataset of 72 individuals aged from 5-96 years (Dadar and Collins, 2019). The BISON script as well as a pretrained random forest classifier is publicly available at http://nist.mni.mcgill.ca/?p=2148.…”

“…Using data from the Single Individual volunteer for Multiple Observations across Networks (SIMON) public dataset, we had a unique opportunity to perform such comparison across 90 scans of this single individual, acquired within the span of seven years on 28 different sites with 12 different scanner models. In this study, we compared the performance of six publicly available, widely used tissue classification methods (Atropos (Avants et al, 2011); BISON (Dadar and Collins, 2019); Classify_Clean (Cocosco et al, 2003); FAST (Zhang et al, 2001); FreeSurfer (Fischl, 2012); and SPM12 (Ashburner et al, 2014, p. 12). These comparisons provide a benchmark on the reliability of each technique, and the amount of variability that can be expected when using multi-center datasets.…”