BioMamba: Leveraging Spectro-Temporal Embedding in Bidirectional Mamba for Enhanced Biosignal Classification

TL;DR

BioMamba tackles biosignal classification by uniting spectral-temporal embeddings with a Bidirectional Mamba backbone and Sparse Feed Forward layers. The approach fuses frequency- and time-domain information, processes embeddings bidirectionally to capture long-range dependencies with linear-like complexity, and reduces parameters via sparsity. Empirical results on six biosignal datasets (EEG and ECG) show state-of-the-art performance on five tasks across six metrics, with substantial reductions in FLOPs and memory usage relative to strong baselines. This work demonstrates a scalable, robust solution for wearable biosignal analysis that balances accuracy and efficiency in resource-constrained settings.

Abstract

Biological signals, such as electroencephalograms (EEGs) and electrocardiograms (ECGs), play a pivotal role in numerous clinical practices, such as diagnosing brain and cardiac arrhythmic diseases. Existing methods for biosignal classification rely on Attention-based frameworks with dense Feed Forward layers, which lead to inefficient learning, high computational overhead, and suboptimal performance. In this work, we introduce BioMamba, a Spectro-Temporal Embedding strategy applied to the Bidirectional Mamba framework with Sparse Feed Forward layers to enable effective learning of biosignal sequences. By integrating these three key components, BioMamba effectively addresses the limitations of existing methods. Extensive experiments demonstrate that BioMamba significantly outperforms state-of-the-art methods with marked improvement in classification performance. The advantages of the proposed BioMamba include (1) Reliability: BioMamba consistently delivers robust results, confirmed across six evaluation metrics. (2) Efficiency: We assess both model and training efficiency, the BioMamba demonstrates computational effectiveness by reducing model size and resource consumption compared to existing approaches. (3) Generality: With the capacity to effectively classify a diverse set of tasks, BioMamba demonstrates adaptability and effectiveness across various domains and applications.

Figures (9)

• Figure 1: Our BioMamba consistently outperforms state-of-the-art biosignals classification methods across six quality evaluation metrics with the average six datasets results.
• Figure 2: The biosignals dimension information and three main types of embeddings.
• Figure 3: An overview of the proposed BioMamba. (a)-(b) Comparison between the Blocks of vanilla Mamaba and the proposed BioMamba. (c) Details of BioMamba blocks: (1). Spectro-Temporal Embedding strategy. (2). Bidirectional Mamba framework. (3). Sparse Feed Forward layers.
• Figure 4: Visualization of the CrowdSource dataset in both the time domain and frequency domain. To enhance the clarity of the channel information, we apply normalization and offset adjustments to the original data.
• Figure 5: The detailed structure of the BioMamba block, Mamba process, and the Selective SSM mechanism.