Automated Huntington's Disease Prognosis via Biomedical Signals and Shallow Machine Learning

Area of Science:

• Biomedical Engineering
• Neuroscience
• Computational Biology

Background:

• Huntington's disease (HD) is a rare, genetic neurodegenerative disorder.
• Current HD prognosis methods are resource-intensive and struggle to differentiate symptomatic from asymptomatic patients.
• Biomedical signal analysis shows promise for diagnosing neurological disorders.

Approach:

• Utilized a certified dataset of electroencephalography (EEG), electrocardiography (ECG), and functional near-infrared spectroscopy (fNIRS) data from HD patients and controls.
• Preprocessed signals and extracted features, applying various shallow machine learning algorithms.
• Employed an Extremely Randomized Trees algorithm for classification.

Key Points:

• The Extremely Randomized Trees model achieved a high accuracy of 91.353% and an AUC of 0.963.
• Feature analysis revealed significant contributions from raw neural and cardiac signals (p<0.05 for 60.865% of features).
• The study highlights the potential of multimodal biomedical signals for HD detection.

Conclusions:

• Neural and cardiac signals are promising biomarkers for detecting HD-related abnormalities.
• This approach can potentially improve early diagnosis and disease progression monitoring in Huntington's disease.
• Machine learning analysis of biomedical signals offers a viable alternative to current complex HD prognosis methods.