Fractal analysis for cognitive impairment classification in DAVF using machine learning.

Area of Science:

• Neuroscience
• Biomarker Discovery
• Medical Imaging Analysis

Background:

• Intracranial dural arteriovenous fistulas (DAVFs) are acquired vascular abnormalities.
• Cognitive impairment is a common symptom in DAVFs, linked to disrupted brain network connectivity.
• Resting-state functional MRI (rsfMRI) is used to study brain connectivity, but fractal patterns in signals complicate analysis.

Purpose of the Study:

• To explore nonfractal connectivity as a potential biomarker for cognitive impairment in DAVF patients.
• To isolate short-memory components in BOLD signals to improve connectivity analysis.
• To differentiate cognitive impairment in DAVF patients using machine learning.

Main Methods:

• 50 DAVF patients and 50 controls underwent neuropsychological assessments and rsfMRI.
• Wavelet transforms decomposed BOLD signals into fractal and nonfractal components.
• Machine learning classifiers (SVM, decision trees) were trained on connectivity matrices for classification.

Main Results:

• Nonfractal connectivity achieved 89.82% accuracy in classifying cognitive impairment using SVM.
• Nonfractal measures outperformed fractal and Pearson correlation methods.
• High sensitivity (86.54%), specificity (92.4%), and AUC (0.96) were obtained for nonfractal connectivity.

Conclusions:

• Nonfractal connectivity shows promise as a biomarker for cognitive impairment in DAVF patients.
• This approach may aid in early diagnosis and intervention for DAVF-related cognitive deficits.
• Further validation with larger datasets is recommended to confirm findings and explore broader applications.