Frequency Bands

Low-gamma was the strongest band in the combined-dataset band experiments. Delta also produced strong subject-level classification performance. Aggregating multiple segment decisions improved subject-level results over segment-level performance. The low-gamma result suggests that Alzheimer’s EEG classification should examine higher-frequency deviations as well as conventional slowing markers. Band-specific experiments showed that sub-band EEG inputs outperformed full-band EEG at the subject level. The paper argues that decomposing EEG into rhythms reduces spectral interference and noise masking from full-band signals.