Machine Learning to Predict Brain Amyloid Pathology in Pre-dementia Alzheimer’s Disease Using QEEG Features and Genetic Algorithm Heuristic

• DOI: 10.3389/fncom.2021.755499
• Publication Date: 2021-11-11T13:55:01Z
• Authors: Nam Heon Kim; Dong Won Yang; Seong Hye Choi; Seung Wan Kang
• Keywords: BETA (programming language); Amyloid (mycology)

The use of positron emission tomography (PET) as the initial or sole biomarker β-amyloid (Aβ) brain pathology may inhibit Alzheimer's disease (AD) drug development and clinical due to cost, access, tolerability. We developed a qEEG-ML algorithm predict Aβ among subjective cognitive decline (SCD) mild impairment (MCI) patients, validated it using PET. compared QEEG data between patients with MCI those SCD without PET-confirmed beta-amyloid plaque. resting-state eyes-closed electroencephalograms (EEG) patterns amyloid positive negative groups relative power measures from 19 channels (Fp1, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, O2), divided into eight frequency bands, delta (1-4 Hz), theta (4-8 alpha 1 (8-10 2 (10-12 beta (12-15 (15-20 3 (20-30 gamma (30-45 Hz) calculated by FFT denoised iSyncBrain®. resulting 152 features were analyzed genetic strategy identify optimal feature combinations maximize classification accuracy. Guided gene modeling methods, we treated each channel band EEG modeled every possible combination within given dimension. then collected models that showed best performance identified genes appeared most frequently in superior models. By repeating this process, converged on model approximates optimum. found average increased iterative progressed. ultimately achieved 85.7% sensitivity, 89.3% specificity, 88.6% accuracy positive/negative classification, 83.3% 84.6% classification.