A novel approach to ASD detection using intuitionistic fuzzy sets and graph convolutional networks
In this work, we propose a novel approach for distinguishing individuals with Autism Spectrum Disorder (ASD) using Intuitionistic Fuzzy Set (IFS) theory and Multi-Scale Enhanced Graph Convolutional Networks (MSE-GCNs). The proposed method extracts features from functional connectivity (FC) fMRI data together with available phenotypic information, including age, sex, clinical scores, and behavioral test results. Each subject is represented as a node in a population graph, where weighted edges are computed on the basis of phenotypic similarity. These edge weights are modeled using IFS-based degrees of membership, non-membership, and hesitancy to capture uncertainty in phenotypic traits. The resulting graph, composed of nodes and IFS-based weighted edges, is then provided as input to the MSE-GCN. This supervised framework captures complex imaging patterns associated with ASD. During testing, the trained network predicts ASD diagnoses by leveraging both the graph structure and fuzzy-enhanced edge relationships. Quantitative results on the ABIDE dataset indicate that the proposed model outperforms existing methods, achieving 89% accuracy and an F1 score of approximately 84%. These findings demonstrate strong classification performance, particularly in real-world settings involving highly imbalanced data. Furthermore, the integration of intuitionistic fuzzy logic into graph-based learning improves the interpretability, reliability, and effectiveness of ASD identification. By accounting for data uncertainty and phenotypic heterogeneity, the proposed system represents a substantial improvement over existing models for ASD and potentially for other neurological disorders.