An anxiety screening framework integrating multimodal data and graph node correlation

Anxiety disorders are a significant global health concern, profoundly impacting patients’ lives and social functioning while imposing considerable burdens on families and economies. However, current anxiety screening methods face limitations due to cost constraints and cognitive biases, particularly in their inability to deeply model correlations among multidimensional features. They often overlook crucial information inherent in their internal couplings, limiting their accuracy and applicability in clinical diagnostics. To address these challenges, we propose an advanced anxiety screening framework that integrates multimodal data, such as physiological, behavioral, audio, and textual, using a Graph Convolutional Network (GCN). While our framework draws upon existing technologies such as GCN, one-dimensional convolutional neural networks, and gated recurrent units, the uniqueness of our framework lies in how these components are combined to capture complex spatiotemporal relationships and correlations among multimodal features. Experimental results demonstrate the framework’s robust performance, achieving an accuracy of 93.48%, Area Under Curve of 94.58%, precision of 90.00%, sensitivity of 81.82%, specificity of 97.14%, F1 score of 85.71%. Notably, the method remains effective even when questionnaire data is unavailable, underscoring its practicality and reliability. This anxiety screening approach provides a new perspective for early identification and intervention of anxiety symptoms, offering a scientific basis for personalized treatment and prevention through the analysis of multimodal data and graph structures.