Enhanced Blood Cell Detection in YOLOv11n Using Gradient Accumulation and Loss Reweighting

Automated blood cell detection is of significant importance in the efficient and accurate diagnosis of hematological diseases. The application of this technology has advanced clinical practice in hematology and improved the speed and accuracy of diagnosis, thereby providing patients with more timely medical intervention. In this study, the YOLOv11n model was optimized by integrating gradient accumulation and loss reweighting techniques to improve its detection performance for blood cells in clinical images. The optimized YOLOv11n model shows an improvement in performance. The mAP50 reached 0.9356, the mAP50-95 was 0.6620, and the precision and recall were better than those of existing methods. The model can effectively address issues such as dense cell distribution, cell overlap, and image artifacts. Therefore, it is highly applicable in real-time clinical applications. The results of the ablation experiment demonstrate that there is a synergistic effect between gradient accumulation and loss reweighting, which can improve detection accuracy without increasing the computational burden. The conclusion indicates that the optimized YOLOv11n model has important application prospects as an automated blood cell detection tool and has the potential to integrate with clinical workflows.