Cell-YOLOv8: A Novel Automated Method for Multi-Scale Cervical Cell Detection

Early screening for cervical cancer is a common preventive strategy that can significantly reduce the risk of the disease through regular examinations. In microscopic images of cervical cells, the changes in cellular characteristics are often subtle, and considerable variability exists among individuals, leading to challenges such as false positives and missed detections. Addressing the issue of detecting multi-scale cells is therefore a complex task. To tackle this challenge, we propose the Cell-YOLOv8 object detection model for effective and accurate detection of multi-scale cervical cells. In this study, we first introduce and design the Convolutional Attention Aggregation Module (CAAM). Second, we incorporate the CARAFE upsam-pling operator and reparameterize its design to reconstruct the backbone and neck networks of Cell-YOLOv8. Finally, we conduct comparisons with state-of-the-art detection methods on two distinct datasets (PQCC and Herlev). The experimental results demonstrate that the Cell-YOLOv8 model outperforms existing methods in the domain of multi-scale cervical cell detection, achieving an average precision (mAP) of 80.2%. This approach provides valuable insights for the development of automated cervical cell detection methods.