Brain network similarity using k-cores

Autism Spectrum Disorder (ASD) is extensively studied by medical practitioners, health researchers, and educators. ASD symptoms appear in early childhood, within the first two years of life, but diagnosing it remains challenging due to its complex and diverse nature. Nevertheless, early diagnosis is crucial for effective intervention. Traditional methods rely on behavioral observations, while modern approaches involve applying machine learning (ML) to brain networks derived from fMRI scans. Limited explainability of these advanced techniques poses a significant challenge in gaining clinicians trust. This paper builds on recent works that design explainable approaches for ASD diagnosis from fMRI data preprocessed as graphs. Our research makes three key contributions. Firstly, we demonstrate that a simple approach based on viewing graphs as tables and using tabular data classifiers can achieve the same performance as state-of-art, explainable graph theoretic methods. Secondly, we provide evidence that by adding higher-order connectivity information as attributes does not improve their performance. Most importantly, we show why classification of brain networks is challenging by demonstrating the similarity between graphs belonging to individuals with ASD and those without, using a novel k-core based approach.