Search

Segmentation of white matter lesions and deep grey matter structures is an important task in the quantification of magnetic resonance imaging in multiple sclerosis. In this paper we explore segmentation solutions based on convolutional neural networks (CNNs) for providing fast, reliable segmentations of lesions and grey-matter structures in multi-modal MR imaging, and the performance of these methods when applied to out-of-centre data. We trained two state-of-the-art fully convolutional CNN architectures on the 2016 MSSEG training dataset, which was annotated by seven independent human raters: a reference implementation of a 3D Unet, and a more recently proposed 3D-to-2D architecture (DeepSCAN). We then retrained those methods on a larger dataset from a single centre, with and without labels for other brain structures. We quantified changes in performance owing to dataset shift, and changes in performance by adding the additional brain-structure labels. We also compared performance with freely available reference methods. Both fully-convolutional CNN methods substantially outperform other approaches in the literature when trained and evaluated in cross-validation on the MSSEG dataset, showing agreement with human raters in the range of human inter-rater variability. Both architectures showed drops in performance when trained on single-centre data and tested on the MSSEG dataset. When trained with the addition of weak anatomical labels derived from Freesurfer, the performance of the 3D Unet degraded, while the performance of the DeepSCAN net improved. Overall, the DeepSCAN network predicting both lesion and anatomical labels was the best-performing network examined.

A correction to this article has been published and is linked from the HTML version of this paper. The error has been fixed in the paper.

Purpose of this study was to investigate a potential correlation between the pattern of cerebral veins (CV) on susceptibility-weighted imaging (SWI) and blood oxygen saturation, as well as preoperative brain injury, in neonates with transposition of the great arteries (TGA). Eleven neonates with TGA underwent MRI preoperatively, including SWI, T1- and T2-weighted scans. Images were retrospectively evaluated and appearance of CV was graded from 0 (normal appearance) to 3 (severe prominent appearance). White matter injuries (WMI) and strokes were analysed. Results were correlated with preductal arterial oxygen saturation. As findings one subject showed a normal CV appearance (grade 0) whereas 10 showed pathological prominent CV (grades 1-3); median 2. Mean oxygen saturation ranged between 67.5% and 89.0% (median 81.0%). CV grade and mean oxygen saturation correlated significantly (p = 0.011). WMI were absent in 5 cases, mild in 4, and moderate in 2 cases. We conclude, that SWI has the potential to be used to estimate the current hypoxic burden on brain tissue in TGA newborns by assessing the prominence of the CV.

Volumetric and morphometric studies have demonstrated structural abnormalities related to chronic epilepsies on a cohort- and population-based level. On a single-patient level, specific patterns of atrophy or cortical reorganization may be widespread and heterogeneous but represent potential targets for further personalized image analysis and surgical therapy. The goal of this study was to compare morphometric data analysis in 37 patients with temporal lobe epilepsies with expert-based image analysis, pre-informed by seizure semiology and ictal scalp EEG. Automated image analysis identified abnormalities exceeding expert-determined structural epileptogenic lesions in 86% of datasets. If EEG lateralization and expert MRI readings were congruent, automated analysis detected abnormalities consistent on a lobar and hemispheric level in 82% of datasets. However, in 25% of patients EEG lateralization and expert readings were inconsistent. Automated analysis localized to the site of resection in 60% of datasets in patients who underwent successful epilepsy surgery. Morphometric abnormalities beyond the mesiotemporal structures contributed to subtype characterisation. We conclude that subject-specific morphometric information is in agreement with expert image analysis and scalp EEG in the majority of cases. However, automated image analysis may provide non-invasive additional information in cases with equivocal radiological and neurophysiological findings.