EEG-Based Classical Framework for Autism Spectrum Disorder Diagnosis
The Core Challenge
Autism Spectrum Disorder (ASD) diagnosis relies heavily on behavioral assessments that are time-intensive, subjective, and prone to diagnostic latency. Automated EEG pattern recognition faces severe challenges due to high inter-subject variability, low signal-to-noise ratios, and non-stationary neural rhythms.
Methodological Contribution
Proposes a CNN-ASD framework that decomposes multi-channel continuous EEG into time-frequency scalograms using Continuous Wavelet Transform (CWT), extracting scale-invariant oscillatory patterns across delta, theta, and alpha bands to achieve rapid, objective diagnosis.
Methodological Pipeline & Tensor Transformations
Multi-Channel EEG Acquisition & Preprocessing
Bandpass filtering (0.5–45 Hz), baseline drift correction, and ICA-driven eye-blink/movement artifact removal.
Continuous Wavelet Transform (CWT)
Computes continuous wavelet scalograms across multiple frequency scales to capture transient neural synchrony.
Deep Convolutional Feature Extraction
Hierarchical 2D CNN layers extract discriminative spatial-temporal signatures from transformed scalograms.
ASD Diagnostic Classification
Softmax output produces calibrated probability metrics distinguishing ASD subjects from neurotypical controls.
Reported Evaluation Results
Scientific Guardrails & Validation Boundaries
Validation on larger, multi-site pediatric cohorts with varying electrode configurations remains necessary prior to clinical deployment.
Translation to Enterprise Production
Translating non-stationary time-series telemetry into computer vision representations (CWT scalograms) demonstrates how classical signal transformations combined with modern 2D CNNs can solve complex temporal diagnostic tasks with sub-second inference.
Access Verified Publication Records
All citations, publisher identifiers, and code references adhere to strict verification standards.