Manual and automated tissue segmentation confirm the impact of thalamus atrophy on cognition in multiple sclerosis: A multicenter study.

BACKGROUND AND RATIONALE: Thalamus atrophy has been linked to cognitive decline in multiple sclerosis (MS) using various segmentation methods. We investigated the consistency of the association between thalamus volume and cognition in MS for two common automated segmentation approaches, as well as fully manual outlining. METHODS: Standardized neuropsychological assessment and 3-Tesla 3D-T1-weighted brain MRI were collected (multi-center) from 57 MS patients and 17 healthy controls. Thalamus segmentations were generated manually and using five automated methods. Agreement between the algorithms and manual outlines was assessed with Bland-Altman plots; linear regression assessed the presence of proportional bias. The effect of segmentation method on the separation of cognitively impaired (CI) and preserved (CP) patients was investigated through Generalized Estimating Equations; associations with cognitive measures were investigated using linear mixed models, for each method and vendor. RESULTS: In smaller thalami, automated methods systematically overestimated volumes compared to manual segmentations [ρ=(-0.42)-(-0.76); p-values < 0.001). All methods significantly distinguished CI from CP MS patients, except manual outlines of the left thalamus (p = 0.23). Poorer global neuropsychological test performance was significantly associated with smaller thalamus volumes bilaterally using all methods. Vendor significantly affected the findings. CONCLUSION: Automated and manual thalamus segmentation consistently demonstrated an association between thalamus atrophy and cognitive impairment in MS. However, a proportional bias in smaller thalami and choice of MRI acquisition system might impact the effect size of these findings.

Reduced accuracy of MRI deep grey matter segmentation in multiple sclerosis: an evaluation of four automated methods against manual reference segmentations in a multi-center cohort.

BACKGROUND: Deep grey matter (DGM) atrophy in multiple sclerosis (MS) and its relation to cognitive and clinical decline requires accurate measurements. MS pathology may deteriorate the performance of automated segmentation methods. Accuracy of DGM segmentation methods is compared between MS and controls, and the relation of performance with lesions and atrophy is studied. METHODS: On images of 21 MS subjects and 11 controls, three raters manually outlined caudate nucleus, putamen and thalamus; outlines were combined by majority voting. FSL-FIRST, FreeSurfer, Geodesic Information Flow and volBrain were evaluated. Performance was evaluated volumetrically (intra-class correlation coefficient (ICC)) and spatially (Dice similarity coefficient (DSC)). Spearman's correlations of DSC with global and local lesion volume, structure of interest volume (ROIV), and normalized brain volume (NBV) were assessed. RESULTS: ICC with manual volumes was mostly good and spatial agreement was high. MS exhibited significantly lower DSC than controls for thalamus and putamen. For some combinations of structure and method, DSC correlated negatively with lesion volume or positively with NBV or ROIV. Lesion-filling did not substantially change segmentations. CONCLUSIONS: Automated methods have impaired performance in patients. Performance generally deteriorated with higher lesion volume and lower NBV and ROIV, suggesting that these may contribute to the impaired performance.

Atypical idiopathic inflammatory demyelinating lesions: prognostic implications and relation to multiple sclerosis.

Atypical lesions of a presumably idiopathic inflammatory demyelinating origin present quite variably and may pose diagnostic problems. The subsequent clinical course is also uncertain. We, therefore, wanted to clarify if atypical idiopathic inflammatory demyelinating lesions (AIIDLs) can be classified according to previously suggested radiologic characteristics and how this classification relates to prognosis. Searching the databases of eight tertiary referral centres we identified 90 adult patients (61 women, 29 men; mean age 34 years) with ≥ 1 AIIDL. We collected their demographic, clinical and magnetic resonance imaging data and obtained follow-up (FU) information on 77 of these patients over a mean duration of 4 years. The AIIDLs presented as a single lesion in 72 (80 %) patients and exhibited an infiltrative (n = 35), megacystic (n = 16), Baló (n = 10) or ring-like (n = 16) lesion appearance in 77 (86 %) patients. Additional multiple sclerosis (MS)-typical lesions existed in 48 (53 %) patients. During FU, a further clinical attack occurred rarely (23-35 % of patients) except for patients with ring-like AIIDLs (62 %). Further attacks were also significantly more often in patients with coexisting MS-typical lesions (41 vs. 10 %, p < 0.005). New AIIDLs developed in six (7 %), and new MS-typical lesions in 29 (42 %) patients. Our findings confirm the previously reported subtypes of AIIDLs. Most types confer a relatively low risk of further clinical attacks, except for ring-like lesions and the combination with MS-typical lesions.

Incorporating domain knowledge into the fuzzy connectedness framework: application to brain lesion volume estimation in multiple sclerosis.

A method for incorporating prior knowledge into the fuzzy connectedness image segmentation framework is presented. This prior knowledge is in the form of probabilistic feature distribution and feature size maps, in a standard anatomical space, and "intensity hints" selected by the user that allow for a skewed distribution of the feature intensity characteristics. The fuzzy affinity between pixels is modified to encapsulate this domain knowledge. The method was tested by using it to segment brain lesions in patients with multiple sclerosis, and the results compared to an established method for lesion outlining based on edge detection and contour following. With the fuzzy connections (FC) method, the user is required to identify each lesion with a mouse click, to provide a set of seed pixels. The algorithm then grows the features from the seeds to define the lesions as a set of objects with fuzzy connectedness above a preset threshold. The FC method gave improved interobserver reproducibility of lesion volumes, and the set of pixels determined to be lesion was more consistent compared to the contouring method. The operator interaction time required to evaluate one subject was reduced from an average of 111 min with contouring to 16 min with the FC method.