<p>Complete blood cell detection holds significant value in clinical diagnostics. Conventional manual microscopy methods suffer from time inefficiency and diagnostic inaccuracies. Existing automated detection approaches remain constrained by high deployment costs and suboptimal accuracy. Although deep learning has introduced powerful paradigms to this field, persistent challenges in detecting overlapping cells and multi-scale objects hinder practical deployment. This study proposes the multi-scale YOLO (MS-YOLO), a blood cell detection model built on YOLOv11, incorporating three key architectural innovations. Specifically, the Multi-scale Dilated Residual Module (MS-DRM) enhances multi-scale discriminability through region and semantic residualization; the Dynamic Cross-path Feature Enhancement Module (DCFEM) strengthens feature representations via a bidirectional hierarchical feature calibration mechanism; and the Light Adaptive-weight Downsampling Module (LADS) improves feature downsampling through adaptive spatial weighting while reducing computational complexity. Experimental results on the Complete-Blood-Count (CBC) benchmark show that MS-YOLO achieves accurate detection of overlapping cells and multi-scale objects, particularly small targets such as platelets, whose AP increased by 4.1%. Overall, it achieves an mAP@50 of 97.4%, outperforming 12 existing models. Further validation on the supplementary leukocyte dataset confirms its robust generalization capability, with accuracy comparable to the specialized leukocyte detection model while remaining substantially lighter. Visualization on challenging samples and unseen data provides qualitative evidence of MS-YOLO’s improved detection and robustness, while heatmap analysis reveals the sources of performance gains. These findings indicate that MS-YOLO achieves an optimal balance between accuracy and efficiency, thereby showing potential for clinical deployment and providing reliable technical support for standardized blood cell quantification.</p>