Validation and Optimization of BIANCA for the Segmentation of Extensive White Matter Hyperintensities

Abstract

White matter hyperintensities (WMH) are a hallmark of small vessel diseases (SVD). Yet, no automated segmentation method is readily and widely used, especially in patients with extensive WMH where lesions are close to the cerebral cortex. BIANCA (Brain Intensity AbNormality Classification Algorithm) is a new fully automated, supervised method for WMH segmentation. In this study, we optimized and compared BIANCA against a reference method with manual editing in a cohort of patients with extensive WMH. This was achieved in two datasets: a clinical protocol with 90 patients having 2-dimensional FLAIR and an advanced protocol with 66 patients having 3-dimensional FLAIR. We first determined simultaneously which input modalities (FLAIR alone or FLAIR + T1) and which training sets were better compared to the reference. Three strategies for the selection of the threshold that is applied to the probabilistic output of BIANCA were then evaluated: chosen at the group level, based on Fazekas score or determined individually. Accuracy of the segmentation was assessed through measures of spatial agreement and volumetric correspondence with respect to reference segmentation. Based on all our tests, we identified multimodal inputs (FLAIR + T1), mixed WMH load training set and individual threshold selection as the best conditions to automatically segment WMH in our cohort. A median Dice similarity index of 0.80 (0.80) and an intraclass correlation coefficient of 0.97 (0.98) were obtained for the clinical (advanced) protocol. However, Bland-Altman plots identified a difference with the reference method that was linearly related to the total burden of WMH. Our results suggest that BIANCA is a reliable and fast segmentation method to extract masks of WMH in patients with extensive lesions.

Performance of three freely available methods for extracting white matter hyperintensities: FreeSurfer, UBO Detector and BIANCA

“…To represent our entire dataset adequately, we selected our training dataset in terms of the WMH loads based on the median values of the Fazekas scores. Ling and colleagues (2018) showed better results with a mixed WMH load training dataset than with a training dataset with only high WMH load as Griffanti and colleagues (2016) suggested in their paper.…”

“…With the global threshold in BIANCA this was not the case. Ling and colleagues (2018) validated BIANCA with different input modalities (FLAIR or FLAIR + T1w), with a cohort of patients with CADASIL using a semi-manually generated gold standard of 10 images per sequence. In their dataset, which contained an extremely high WMH load, they received a median DSC of 0.79 (our median 2D FLAIR = 0.560) for the 2D FLAIR + T1w images and a median DSC of 0.78 (our median DSC 3D FLAIR = 0.615) for the 3D FLAIR + T1w images.…”

Automatic Segmentation of White Matter Hyperintensities from Brain Magnetic Resonance Images in the Era of Deep Learning and Big Data – A Systematic Review

“…Nine of them also used additional datasets or patient data from clinics. Out of 37, twelve studies reported using data from prospective studies or clinics (Atlason et al, 2019;Bowles et al, 2017;Guerrero et al, 2017;Hong et al, 2020;Moeskops et al, 2018;Ling et al, 2018;Park et al, 2018;Qin et al, 2018;Rincón et al, 2017;Roy et al, 2015;Sundaresan et al, 2019;Wang et al, 2015). Four studies used data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) (Sudre et al, 2017;Dadar et al, 2017a;Rachmadi et al, 2018Rachmadi et al, , 2020, of which only two declared the subset used (Rachmadi et al, 2018(Rachmadi et al, , 2020.…”

“…Twelve studies used data only acquired at 1.5T, and twelve used data only acquired at 3T. Only 11/37 studies used data acquired at both 1.5T and 3T (Atlason et al, 2019;Dadar et al, 2017a;Jiang et al, 2018;Knight et al, 2018;Li et al, 2018;Ling et al, 2018;Liu et al, 2020;Manjón et al, 2018;Sundaresan et al, 2019;Wang et al, 2015;Wu, Zhang, Wang & Tang, 2019).…”

“…Only 8/37 studies made the code publicly available (Griffanti et al, 2016;Hong et al, 2020;Jiang et al, 2018;Li et al, 2018;Park et al, 2018;Rachmadi et al, 2018Rachmadi et al, , 2020Valverde et al, 2017). A study (Ling et al, 2018) evaluates different configurations of the method described by Griffanti et al (2016) making recommendations of its use. We present risk of bias assessment of the 37 included studies using QUADAS 2 tool in Table 3.…”