An automatic identification method of vital sign parameters based on YOLOv10 and Shape-IoU

Accurate recognition of vital sign parameters from multi-parameter monitors—such as ECG waveforms, blood pressure , and oxygen saturation —is crucial in medical monitoring. However, existing object detection algorithms face performance bottlenecks when dealing with small targets, complex shapes, and noise interference. This paper proposes an end-to-end automatic recognition method based on YOLOv10 and combined with Shape-IoU metric, aiming to enhance detection accuracy and robustness, especially in medical image applications.The proposed method was tested on a real medical dataset of 3,396 labeled images. Shape-IoU was introduced as a new bounding box matching method on the basis of the YOLOv10 framework, comprehensively considering the shape and scale information of the target to optimize detection accuracy. By designing a multi-task loss function, improving feature fusion strategies, and enhancing the training dataset, the model's detection capabilities for small and complex-shaped targets were enhanced.The YOLOv10 combined with Shape-IoU model achieved an mAP of 97.8%, an improvement of 0.025% compared to the baseline model. Compared with traditional models such as ResNet, the mAP was improved by an average of 17 percentage points. The real-time processing capability of the model on an RTX 3090 GPU reached 45 FPS, meeting the requirements of clinical real-time monitoring. Additionally, the false detection rate was 12.3% lower than traditional methods, demonstrating strong anti-noise capabilities.The automatic recognition method based on YOLOv10 and ShapeIoU proposed in this paper significantly improves detection accuracy and real-time performance in vital sign parameter detection, especially in handling small and complex-shaped targets. It provides an efficient and reliable solution for medical image analysis.