YOLO-VML: An Improved Object Detection Model for Blastomeres and Pronuclei Localization in IoMT

Goal: Blastomeres and pronuclei detection plays a crucial role in advancing research on embryo development and assisted reproductive technologies. However, due to the frequent overlapping of blastomeres and the pronuclei's small size, background similarity, and unclear boundaries, their localization proves to be extremely difficult. Methods: To address these challenges, we propose YOLO-VML, an improved detection model based on the YOLOv10 framework. The model integrates the visual state space (VSS) module of VMamba into the backbone network to enhance the global receptive field and enable broader feature capture. A multi-branch weighted feature pyramid network (MBFPN) is introduced as the neck structure to improve the preservation and fusion of features, especially those related to small targets. Additionally, a lightweight shared convolutional detection head (LSCD) is employed to reduce parameters and computational overhead while maintaining detection accuracy. Results: The proposed YOLO-VML model demonstrates excellent performance in detecting both blastomeres and pronuclei. It achieves a mean average precision (mAP@0.5) of 93.2% for pronuclei detection and 92.3% for blastomere detection beyond the 4-cell stage. Conclusions: YOLO-VML effectively addresses the difficulties in blastomere and pronucleus localization by enhancing feature representation and detection efficiency. Its high accuracy and efficiency make it a valuable tool for advancing embryo research and assisted reproductive technology applications.