MobileMamba-HC: Medical Image Disease Detection in Healthcare Integrating Frequency Adaptive Dilated Convolution and Spatial-Channel Synergistic Attention

MobileMamba-HC: Medical Image Disease Detection in Healthcare Integrating Frequency Adaptive Dilated Convolution and Spatial-Channel Synergistic Attention

To address the challenges in disease detection tasks within the medical and healthcare domain, particularly those associated with biopharmaceutical applications, such as insufficient feature representation, limited capability for multi-scale lesion recognition, and severe interference from complex backgrounds, this paper proposes a detection model named MobileMamba-HC, which integrates frequency-domain adaptive convolution with a spatial-channel collaborative attention mechanism. The proposed method aims to improve the perception of fine-grained lesion regions while maintaining model efficiency, thereby enabling more accurate and robust disease detection in medical images. Built upon the MobileMamba architecture, the proposed model fully exploits its strengths in long-sequence modeling and global dependency capture to effectively model long-range spatial relationships in medical images. On this basis, a Frequency-domain Adaptive Dilated Convolution (FADC) module is introduced. By dynamically modulating feature responses in the frequency domain, this module achieves adaptive perception of components at different scales and frequencies, thereby enhancing the model’s ability to represent multi-scale lesion structures. Meanwhile, a Spatial and Channel Synergistic Attention (SCSA) mechanism is designed to jointly model critical regions and discriminative features from both the spatial and channel dimensions, suppressing redundant information and background noise interference and further improving the discriminability of feature representations. Through the synergistic effect of these three components, the proposed model significantly strengthens its capacity for modeling complex medical images while preserving its lightweight characteristics. In the experimental section, extensive evaluations are conducted on multiple public medical imaging datasets, and comparative experiments are performed against various mainstream deep learning detection models. The experimental results demonstrate that the proposed MobileMamba-HC achieves superior performance over the compared methods across multiple evaluation metrics. In particular, it exhibits stronger robustness in small lesion detection and low-contrast scenarios. Furthermore, ablation studies verify the effectiveness and complementarity of the FADC module and theSCSA mechanism in improving model performance, confirming the contribution of each proposed component to the overall enhancement. These findings validate that the proposed MobileMamba-HC model achieves a favorable balance between performance and efficiency in medical image disease detection tasks, providing an effective solution for intelligent computer-aided diagnosis in complex medical scenarios.