MedGNN: Towards Multi-resolution Spatiotemporal Graph Learning for Medical Time Series Classification

TL;DR

The paper tackles medical time series classification by modeling dynamic, multi-channel, multi-resolution spatiotemporal dependencies. It introduces MedGNN, combining adaptive multi-resolution graphs, Difference Attention Networks to mitigate baseline wander, Frequency Convolution Networks for multi-view temporal representations, and a Multi-resolution Graph Transformer to fuse information across scales. Empirical results on five EEG/ECG datasets show MedGNN achieving state-of-the-art performance and robust improvements in ablations, with clear gains from MRGL, FCN, and DA components. The approach holds promise for improved clinical monitoring and early diagnosis by accurately capturing complex physiological patterns across multiple temporal and spectral views.

Abstract

Medical time series has been playing a vital role in real-world healthcare systems as valuable information in monitoring health conditions of patients. Accurate classification for medical time series, e.g., Electrocardiography (ECG) signals, can help for early detection and diagnosis. Traditional methods towards medical time series classification rely on handcrafted feature extraction and statistical methods; with the recent advancement of artificial intelligence, the machine learning and deep learning methods have become more popular. However, existing methods often fail to fully model the complex spatial dynamics under different scales, which ignore the dynamic multi-resolution spatial and temporal joint inter-dependencies. Moreover, they are less likely to consider the special baseline wander problem as well as the multi-view characteristics of medical time series, which largely hinders their prediction performance. To address these limitations, we propose a Multi-resolution Spatiotemporal Graph Learning framework, MedGNN, for medical time series classification. Specifically, we first propose to construct multi-resolution adaptive graph structures to learn dynamic multi-scale embeddings. Then, to address the baseline wander problem, we propose Difference Attention Networks to operate self-attention mechanisms on the finite difference for temporal modeling. Moreover, to learn the multi-view characteristics, we utilize the Frequency Convolution Networks to capture complementary information of medical time series from the frequency domain. In addition, we introduce the Multi-resolution Graph Transformer architecture to model the dynamic dependencies and fuse the information from different resolutions. Finally, we have conducted extensive experiments on multiple medical real-world datasets that demonstrate the superior performance of our method. Our Code is available.

Figures (8)

• Figure 1: The overall architecture of MedGNN. (a) Multi-resolution graph construction is utilized to learn the dynamic spatiotemporal representations. (b) Frequency convolution networks are applied to provide a multi-view perspective of the temporal dynamics by applying convolutions in the frequency domain. (c) Difference attention networks are employed to capture key temporal patterns while mitigating the impact of baseline wander. (d) Multi-resolution graph transformer is leveraged to model the dynamic spatial dependencies and and fuse the information from different resolutions.
• Figure 2: Ablation study of Multi-Resolution Graph Learning (MRGL) under the Subject-based setup.
• Figure 3: Ablation study of Frequency Convolution Networks (FCN) under the Subject-based setup.
• Figure 4: Model effectiveness and efficiency comparison on two datasets under the Subject-based setup.
• Figure 5: Adjacent matrices of multi-resolution graphs learned from APAVA dataset.