Author Of 2 Presentations
P0559 - Cluster Analysis discriminates Multiple Sclerosis Patients based on Lesion Size and Myelin Content (ID 1884)
Abstract
Background
Clinical heterogeneity among patients with multiple sclerosis (MS) may be driven by genetic and environmental influences that lead to distinctive MRI features.
Objectives
Our objective was to utilize a cluster analysis to determine the variability of quantitative MRI features among a cohort of MS patients and examine the ability of these imaging features to discriminate patients by clinical disability.
Methods
Ninety-six relapsing remitting MS patients and 7 patients with progressive MS underwent Fast Acquisition with Spiral Trajectory and T2prep (FAST-T2) sequence, for myelin water fraction (MWF) analysis, and conventional MRI for measures of lesion volume, cortical thickness and thalamic volume. An agglomerative hierarchical clustering algorithm was implemented using lesion level MRI features selected from a Principal Component Analysis (PCA). The final clusters were selected by implementing a comprehensive validation method based on several unsupervised statistical learning techniques. Matched cluster groups with statistically significant clinical covariates (i.e. age and disease duration) were analyzed based on propensity scores.
Results
A total of 1691 chronic MS lesions were identified among the 103 MS patients. Mean patient age was 44.4 (+/- 11.9) years, disease duration was 10.5 (+/- 8.3) years, and expanded disability status scale (EDSS) was 2.2 (+/- 2.0). PCA demonstrated lesion MWF and volume distributions characterized by 25th, 50th and 75th percentiles account for 87% of the total variability. The hierarchical clustering confirmed two distinct patient clusters. The variables in order of importance were individual lesion median MWF, MWF 25th, MWF 75th, volume 75th percentiles, median individual lesion volume, and total lesion volume (all p-values < 0.000001). Cortical thickness and thalamic volume were significant but less important on cluster discrimination. The clustering MRI features discriminated patients based upon EDSS, p=0.0016 at the time of MRI and maintained EDSS difference at five years (n=72), p=0.0016.
Conclusions
The size and extent of demyelination among individual lesions discriminated MS patients into two MRI lesion-based clusters and was associated with clinical disability. These results suggest an inherent difference among patients with regard to lesion pathology and repair.
P0585 - Hurst exponent as an imaging biomarker of impairment in multiple sclerosis (ID 1426)
Abstract
Background
Hurst Exponent (HE) is a scalar measurement of long-term temporal memory of a time series. The HE has been found in previous studies to be an effective means of measuring the long-range temporal dependence of brain activity as measured by fMRI. We hypothesize that the HE can be associated with impairment in Multiple Sclerosis (MS), therefore can be an imaging biomarker of MS.
Objectives
The primary goal of the study is to assess how well HE measurements can distinguish MS patients from healthy controls (HC), as well as MS patients with impairment from those without impairment. The second objective was to identify which brain regions’ HE alterations are associated with impairment in MS.
Methods
Fifteen HC (age: 43.66±8.64, 53% female) and 76 MS patients (age:45.28±11.46 years, 65% female, disease duration:12.29±7.25 years) were included in our study; 23 had EDSS2 at study baseline. Logistic ridge regression (LR) was used to classify two groups: (1) HC vs MS patients and (2) MS patients with vs without impairment. The classification tasks were performed using HE measurements in 86 cortical and subcortical regions. Five-fold cross-validation was used to train, validate, and test the model, with 10 outer loop repetitions. Area Under ROC curve (AUC) over the folds was used to assess classification performance.
Results
HE was found to be significantly higher in the non-impaired group compared to the impaired group in the right superior frontal gyrus (corrected p-value=0.025). The classification of HC vs MS had an AUC of 0.65 (IQR:0.18), while the task of classifying MS patients by impairment level had an AUC of 0.63 (IQR: 0.12). For the classification of HC vs MS, the regions that were the most predictive were in deep gray matter. Lower HE in the left amygdala, left thalamus and left putamen, and higher HE in the left hippocampus was associated with MS. For the classification of impairment level in MS, deep gray matter regions were also important, as were HE in the frontal lobe. Lower HE in the left caudate, right and left amygdala, and left superior frontal and higher HE in the right ventral DC were associated with more impairment in MS.
Conclusions
HE was found to be moderately discriminative between HC and MS and within impairment levels in MS. HE in subcortical and superior frontal regions were found to be important biomarkers of impairment severity in MS, as well as in distinguishing MS patients from HC. Further research is necessary to identify the mechanism driving these differences.