MedMamba: Multi-View State Space Models with Adaptive Graph Learning for Medical Time Series Classification

2026-05-24

AI Summary

This work proposes MedMamba, an end-to-end architecture that addresses key limitations of existing methods for medical time series analysis—namely, their inability to jointly model local and global dynamics, handle non-stationarities such as baseline drift, and effectively capture sparse, directed inter-channel dependencies. MedMamba uniquely integrates multi-view state space models with adaptive graph learning: it employs multi-scale convolutions to extract local features and a tri-branch differential state space encoder to concurrently model raw, time-domain differential, and frequency-domain views, thereby mitigating non-stationary interference. Furthermore, it introduces a learnable sparse directed graph structure to capture channel dependencies without requiring a predefined graph. Evaluated on five real-world medical time series datasets, MedMamba achieves state-of-the-art performance while maintaining linear computational complexity, and ablation studies confirm the contribution of each component.

Abstract

Medical time series are central to healthcare, enabling continuous monitoring and supporting timely clinical decisions. Despite recent progress, existing methods struggle to jointly model local-global dynamics and handle nonstationarities like baseline drift, while often failing to capture latent channel interactions. To address these challenges, we propose MedMamba, an end-to-end architecture that integrates state space models with domain-specific inductive biases. Specifically, MedMamba first employs multi-scale convolutional embeddings to capture discriminative local morphology. Second, to mitigate nonstationarity, we introduce a tri-branch differential state space encoder that processes raw, temporal-difference, and frequency-domain views, fusing them to emphasize informative patterns while suppressing drift. Furthermore, to uncover latent channel correlations, we design a spatial graph Mamba module that learns a directed dependency structure regularized toward sparsity and acyclicity, which obviates the need for predefined graphs. Extensive experiments on five real-world datasets demonstrate that MedMamba achieves state-of-the-art performance while maintaining linear computational complexity, and ablation studies validate each component's contribution.Code is available at https://github.com/zhangda1018/MedMamba.

Problem

• medical time series
• nonstationarity
• local-global dynamics
• channel interactions
• time series classification

Innovation

• State Space Models
• Adaptive Graph Learning
• Medical Time Series Classification
• Nonstationarity Handling
• Multi-View Fusion