Machine Learning Applied to EEG Data May Serve as Huntington’s Biomarker, Study Suggests
The research, “EEG may serve as a biomarker in Huntington’s disease using machine learning automatic classification,” appeared in the journal Scientific Reports.
Finding reliable markers of disease progression is a current goal in Huntington’s research; qEEG, a technique that measures electrical activity in the brain, is an inexpensive and rapid way to analyze brain alterations occurring before or in parallel to motor or cognitive changes in patients. Combining this information with clinical tests in gene carriers may enable greater insights on disease progression as well as increased sensitivity to detect subtle changes.
Because previous studies had found abnormal EEG data in patients with Huntington’s, researchers hypothesized that using machine learning — a subset of artificial intelligence — for automatic classification of EEG patterns could enable differentiation of Huntington’s gene carriers from healthy controls.
The investigators also intended to find qEEG features that correlate with frequently used clinical and cognitive markers in people with Huntington’s.
A total of 26 Huntington’s gene carriers (with a mean age of 49.7 years) and 25 healthy controls (mean age of 52.7 years) were recruited from the Leiden University Medical Center, in the Netherlands. Three of the controls were family members of recruited patients.
Six of the gene carriers did not have any disease symptoms (pre-Huntington’s). These participants and the 20 with early Huntington’s had a minimum of 40 CAG repeats —the typical HTT gene alterations in Huntington’s disease. Patients with early manifest disease also had a Total Functional Capacity (TFC) score of 7 or higher (range is from 0 to 13, with higher values indicating greater capacity).
The team also used the Unified Huntington’s Disease Rating Scale (UHDRS-TMS) — which assesses motor and cognitive functions, behavioral abnormalities, and functional capacity — the neurocognitive Symbol Digit Modalities Test (SDMT) and the Stroop Word Reading (SWR) test, as well as the Beck Depression Inventory-II (BDI-II), a measure of depression severity.
EEG was recorded over three minutes with the participants awake, but at rest with their eyes closed. TFC, SWR and UHDRS-TMS scores were significantly worse in Huntington’s gene carriers than in controls. Although not statistically significant, the carriers showed a trend for worse SDMT and BDI-II scores. As for the carrier subgroups, the group with early manifest Huntington’s had worse SDMT scores than controls only, and showed more severe depression than both pre-Huntington’s and control participants.
An EEG index (0-1) was then created by identifying statistical patterns in a large set of EEG data. A low index value indicated a state close to normal, while a high value suggested Huntington’s. It had a specificity, sensitivity and accuracy of 83% to separate Huntington’s gene carriers from controls. The two carrier subgroups were pooled together in the EEG analysis, given the low number of pre-Huntington’s participants, the team noted.
In Huntington’s gene carriers, specific qEEG features had highly significant correlations with SDMT and UHDRS-TMS scores, the scientists found.
“In this exploratory study we show promising results where qEEG related modalities may help to unravel how [Huntington’s] evolves,” they wrote. “The results suggest that qEEG may serve as a biomarker in HD.”
However, the team cautioned that as its study is exploratory, the findings need to be confirmed in larger longitudinal studies.