Spectral feature modeling with graph signal processing for brain connectivity in autism spectrum disorder

Abstract

Autism spectrum disorder (ASD) is a complex neurodevelopmental condition associated with disrupted brain connectivity. Traditional graph-theoretical approaches have been widely employed to study ASD biomarkers; however, these methods are often limited to static topological measures and lack the capacity to capture spectral characteristics of brain activity, especially in multimodal data settings. This limits their ability to model dynamic neural interactions and reduces their diagnostic precision. To overcome these limitations, we propose a Graph Signal Processing (GSP)-based framework that integrates spectral-domain features with topological descriptors to model brain connectivity more comprehensively. Using publicly available fMRI and EEG datasets, we construct subject-specific connectivity graphs where nodes represent brain regions and edges encode functional interactions. We extract advanced GSP features such as Graph Fourier Transform coefficients, spectral entropy, and clustering coefficients, and combine them using Principal Component Analysis (PCA). These are classified using a Support Vector Machine (SVM) with a radial basis function (RBF) kernel. The proposed model achieves 98.8% classification accuracy, significantly outperforming prior multimodal GSP studies. Feature ablation analysis reveals that spectral entropy contributes most to this improvement, with its removal resulting in a nearly 30% performance drop. Additionally, a 25% sparsity threshold in graph construction was found to maximize both robustness and computational efficiency. These findings demonstrate that incorporating frequency-domain information through GSP enables a more discriminative and biologically meaningful representation of ASD-related neural patterns, offering a promising direction for accurate diagnosis and biomarker discovery.

Keywords: Autism spectrum disorder; Brain connectivity analysis; Graph Fourier transform; Graph signal processing; Spectral clustering.

Fig. 1

Axial fMRI brain slices illustrating neural activation patterns associated with five emotional states: Afraid, Calm, Delighted, Depressed, and Excited. Activation regions are highlighted in orange-yellow, showing distinct patterns of neural engagement for each emotional condition. These images underscore the differential functional connectivity and brain region involvement tied to emotional processing.

Fig. 2

Overall architecture of proposed methodology.

Fig. 3

The diagram illustrates the comprehensive methodology framework for preprocessing multimodal data, including Resting-State fMRI and EEG, for brain connectivity analysis.

Algorithm 1

Spectral clustering.

Algorithm 2

Algorithm for ASD classification using GSP and SVM.

Fig. 4

This figure visualizes the brain connectivity for the “Afraid” emotional state. Subplots include node intensity visualization, connectivity labels, highlighted high-intensity regions, and graph legend representation. The “Afraid” state shows increased activity in specific areas, suggesting heightened brain activity linked to fear processing.

Fig. 5

This figure illustrates the brain connectivity in a “Calm” state. The graph’s balanced connectivity suggests a more stable distribution of activity, as evidenced by fewer high-intensity regions compared to the “Afraid” state. Subplots help to differentiate key nodes and the flow of information across different areas of the brain.

Fig. 6

The figure represents the “Delighted” state, highlighting brain regions associated with positive emotional response. High-intensity nodes are primarily clustered at the center, indicating strong activation in specific areas associated with positive emotions.

Fig. 7

The “Excited” state shows distinct high-intensity regions, which suggest a significant level of activation in specific areas of the brain related to arousal. The graph also highlights multiple strongly connected nodes, indicative of the increased overall activity during excitement.

Fig. 8

This figure represents brain activity in the “Depressed” state. High-intensity regions are more spread out compared to other states, reflecting the changes in neural activation that may correspond to depressive symptoms. Detailed connectivity analysis provides insights into the patterns of connectivity characteristic of this state.

Fig. 9

This spectrum represents the eigenvalues of the graph Laplacian for the “Afraid” state. The gradual increase in eigenvalues indicates a high degree of clustering, particularly highlighting areas of the brain that are highly connected during fearful experiences.

Fig. 10

The spectrum shows a relatively smooth progression, suggesting a well-distributed connectivity pattern in the “Calm” state. The lack of sudden changes in eigenvalues indicates fewer distinct clusters, reflecting a more balanced state.

Fig. 11

The “Delighted” state spectrum reveals increased clustering in specific regions, shown by the presence of lower eigenvalues. This suggests enhanced connectivity in areas associated with positive emotional response.

Fig. 12

The “Depressed” state exhibits variability in its eigenvalues, which suggests irregular connectivity and possible disconnection between brain regions. This may represent the altered neural processing in individuals experiencing depressive symptoms.

Fig. 13

The “Excited” state shows significant variations in the Laplacian spectrum, indicating dynamic connectivity. The spectrum suggests a complex interplay of various brain regions, likely corresponding to the heightened arousal of an excited state.

Fig. 14

Random forest feature importance showing that Spectral Entropy and Clustering Coefficients are significant for ASD classification, indicating complex signal variability and abnormal clustering in ASD.

Fig. 15

SVM feature importance showing similar patterns to Random Forest, with notable emphasis on Spectral Entropy and Average Path Length, indicating signal complexity and network efficiency as crucial factors in ASD classification.

Fig. 16

k-NN feature importance highlighting clustering coefficient and spectral entropy, emphasizing the significance of local network patterns and temporal brain dynamics in ASD classification.

Fig. 17

EEG signal—raw vs. filtered: the filtered signal reveals clear, structured patterns essential for accurate feature extraction, highlighting the efficacy of preprocessing.

Fig. 18

Frequency spectrum of EEG, showing the dominance of Alpha and Gamma bands, which are linked to attention and cognitive processing, often disrupted in ASD.

Fig. 19

KDE plots for feature distributions showing the differences in “spectral entropy” and “path length variability” indicate the heightened complexity and network irregularity in ASD.