Frequency-Aware Feature Fusion Driven Multimodal Cell Microscopic Image Segmentation Framework

Multimodal cell microscopic image segmentation is a core component of high-content imaging and analysis (HCIA) technology, and its segmentation accuracy directly impacts the precision of HCIA analysis results. Deep learning-based cell segmentation methods have been widely applied due to their end-to-end nature. However, when processing multimodal cell microscopy images, challenges such as missed cell detection, image degradation, and insufficient feature utilization persist, leading to limited segmentation accuracy. To address these issues, this paper proposes a novel deep learning framework designed to achieve accurate and efficient segmentation of multimodal cell microscopy images without manual parameter tuning or algorithm switching. This framework enhances segmentation accuracy through three core modules: a weighted bidirectional feature pyramid network (BiFPN), which reduces the problem of missed cell detection in low-contrast regions by constructing a weighted bidirectional cross-scale connection mechanism; frequency-aware feature fusion (FreqFusion), which precisely identifies cell boundaries under complex degradation conditions through a frequency-domain adaptive mechanism; and a mixed local channel attention (MLCA) mechanism, which guides the model to focus on critical channels and regions that are difficult to segment. On a custom dataset, the proposed framework achieved an average precision of 95.07%, a cell detection rate of 96.72%, and a segmentation speed of 59.23 FPS. Its robust generalization ability and computational efficiency were further validated on public datasets. These advancements lay a solid foundation for the quantitative analysis of microscopic images in precision medicine.