A Quantum-Based Machine Learning Approach for Autism Detection Using Common Spatial Patterns of EEG Signals

Autism Spectrum Disorder (ASD) significantly impacts social communication, interaction, and behavior. Early diagnosis and timely intervention can improve outcomes by enabling tailored therapeutic strategies. Electroencephalography (EEG) has emerged as a non-invasive tool to capture brain activity and facilitate the early detection of ASD using machine learning techniques. However, attaining high accuracy with minimal EEG channels remains a challenge. This study analyzed EEG data from 10 children with ASD and 10 Typically Developed (TD) children using three electrode combinations: C3-C4, C3-Cz, and C4-Cz. EEG signals were spatially filtered using a wavelet-based regularized filter bank common spatial pattern. Key features, including peak-to-peak amplitude, were extracted, and correlation-based feature selection identified the most informative features. Classification with Support Vector Machine (SVM) identified the C4-Cz pair as the most effective, achieving the highest accuracy. Further analysis applied Neural Networks (NN), Quantum Support Vector Machines (QSVM), and Quantum Neural Networks (QNN) to classify data from the C4-Cz pair. QSVM with amplitude embedding feature map outperformed others, achieving an accuracy of 94.7%. Performance was further improved by incorporating an enhanced feature set comprising peak frequency, Stockwell transform coefficients, and peak-to-peak amplitude. The proposed system, leveraging these refined features and QSVM, achieved an exceptional accuracy of 98.9%. To our knowledge, this is the first study utilizing an enhanced feature set derived from reduced brain lobes and quantum machine learning for ASD classification, offering a novel and highly accurate diagnostic approach.