Quasi-Circular Cell Detection via Frequency-Enhanced and Shape-Aware Features

Cell detection in medical microscopy images remains a challenging task due to irregular shapes, blurred boundaries, and significant scale variations. To address these challenges, particularly in detecting roundish object such as white blood cells and cervical cells, this paper introduces a novel detection framework that combines frequency-enhanced feature extraction with multi-scale attention fusion, specifically designed for circular object detection. The framework incorporates a Frequency-Spatial Fusion Module (FSFM), which applies FFT-based high-pass filtering to extract fine-grained high-frequency details and utilizes frequency-aware attention mechanisms to enhance the integration of multi-scale features. In addition, Circular-Region Perception Module (CRPM) are embedded into the backbone network, enabling early-stage representation of quasi-circular structures commonly found in cytological imagery. This design helps preserve shape continuity and structural integrity throughout the feature hierarchy. Extensive evaluations on benchmark cytology datasets show that the proposed method has higher accuracy and robustness than traditional baseline methods. These results highlight its effectiveness and practicality for real-world medical image analysis.