Wavelet Based fMRI Analysis for Autism Spectrum Disorder Detection using Feature Selection and Ridge Classifier

As autism spectrum disorder (ASD) rates rise, timely diagnosis and treatment are increasingly crucial. However, current diagnostic methods rely on subjective criteria, such as clinical observations and tests, which are costly and time-consuming. Functional Magnetic Resonance Imaging (fMRI) shows promising potential in identifying ASD. However, its use is limited by cost and data availability issues. Leveraging machine learning (ML) can help extract meaningful features, potentially enhancing ASD diagnosis. This paper presents an ML-based framework for identifying ASD using wavelet transform, Tangent Pearson (TP) embedding, Principal Component Analysis (PCA), Analysis of Variance (ANOVA) feature selection method, and Maximum Independence Domain Adaptation (MIDA) algorithm. Wavelet transform extracts different frequency levels of the input Blood Oxygen Level-dependent signals. Subsequently, the signals undergo processing via the TP embedding algorithm, followed by PCA and ANOVA for feature reduction and selection. Due to the different types of fMRI scanning inducing domain shift, MIDA is applied to align feature representation. This alignment is in such a way that they become maximally independent while preserving relevant information for classification tasks. The achieved state-of-the-art Area Under the Curve metric stands at 79.01, with an accuracy rate of 72.47%.