Search papers, labs, and topics across Lattice.
This study introduces a Multi-View Gated Graph Attention Network that leverages Automatic Speech Recognition to analyze spontaneous speech for Alzheimer's Disease detection by constructing semantic, dependency, and co-occurrence graphs. The innovative use of Pointwise Mutual Information (PMI) allows the model to quantify narrative logic and linguistic deviations, addressing the non-linear structural disruptions and clinical heterogeneity often overlooked in existing systems. Evaluated on the ADReSSo dataset, the model achieves an impressive 90.00% accuracy, with ablation studies highlighting the critical role of the PMI-based graph and adaptive gating in enhancing classification robustness across diverse clinical populations.
Achieving 90% accuracy in Alzheimer's detection through a novel graph-based approach that captures the intricacies of spontaneous speech could revolutionize diagnostic practices.
Spontaneous speech is a vital non-invasive biomarker for Alzheimer's Disease (AD), yet many systems overlook non-linear structural disruptions and clinical heterogeneity in pathological language. We propose a Multi-View Gated Graph Attention Network that transcribes audio via Automatic Speech Recognition (ASR) to construct semantic, dependency, and co-occurrence graphs, characterizing speech through a "content-structure-flow" framework. Notably, the co-occurrence graph leverages Pointwise Mutual Information (PMI) from a normative corpus to quantify narrative logic and linguistic deviation. To address symptomatic diversity, an adaptive gated fusion mechanism dynamically integrates these views. Evaluated on the ADReSSo dataset, our model achieves 90.00% accuracy. Ablation results confirm that the PMI-based graph and heterogeneity-aware gating are essential for robust classification across diverse clinical populations. Our source code is publicly available at https://github.com/opeacc/AD.
