Objective: To identify a robust imaging marker that is capable of predicting real life disease onset in Huntington’s disease (HD).

Background: Huntington’s disease (HD) is an incurable, inherited, progressive, neurodegenerative disorder that is characterised by a triad of motor, cognitive and psychiatric problems. While it is possible to reliably identify individuals who carry the expanded HD gene, at present there is no robust way of determining when gene carriers will develop clinical disease. Perfecting this prediction is a necessary prerequisite in the search for neuroprotective therapies in HD.

Methods: Resting state fMRI and structural MRI scans were conducted on 19 pre-manifest HD gene carriers (pre-HD) and 21 healthy controls and resting state network couplings (RSNCs), subcortical grey matter volumes (SV) and cortical thicknesses (CT) were measured. A multivariate machine learning approach (a support vector machine, SVM) was applied to the data to determine the accuracy of each measure, individually and as a combination, at correctly classifying pre-manifest HDs vs. controls; this was then related to proximity to disease onset. Since the scans were conducted 8 pre-HD have been diagnosed with manifest disease (converter). Therefore, the individual measures and SVM model were further examined in terms of accuracy distinguishing unexpected converters from non-converted pre-HD individuals.

Results: The SVM identified converters with an above chance level of accuracy for all analyses (RSN network coupling correct = 72%, accuracy > 95% of permutations; SV correct = 83%, accuracy > 99% of permutations; CT correct = 76%, accuracy > 95% of permutations; all combined correct = 90%, accuracy = all permutations). In addition, distance to the SVM classification hyperplane (a measure of classification strength) significantly differed between the unexpected phenoconverted and the non-phenoconverted preHD individuals (t=3.733, p=0.003 two tailed). Furthermore, contrasting all phenoconverted individuals vs. non-phenoconverted provided a robust cross group difference (t=4.265, p<0.001 two tailed).

Conclusions: By combining both structural and functional brain features with machine learning, we were able to provide a polymarker that predicts the onset of HD symptoms with greater accuracy than the currently applied statistical models.

« Back to 2016 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/combined-imaging-markers-increase-accuracy-when-predicting-real-life-disease-onset-in-huntingtons-disease/