A novel RSCA-YOLOv8s network for automatic diagnosis and graduation in pressure injury

To enhance the accuracy and objectivity in PI diagnosis, this study proposes an improved PI recognition method based on YOLOv8s, which introduces a spatial and channel synergistic attention mechanism in the C2f module to enhance the feature extraction capability and embeds a multi-scale fusion module to improve the model's ability to recognize PI varying scales. This study was conducted from January 2024 to December 2024, during which 366 PI images were collected by standardized trained nurses from two tertiary Grade A hospitals in Jiaxing. The dataset was divided into a training set and a validation set in an 8:2 ratio. The improved YOLOv8s, YOLOv5, TPH-YOLO, YOLOv7, YOLOv8s, and Swin Transformer models were employed for training. Model performance was evaluated using precision, recall, F1-score, mean average precision(mAP₅₀), and mean average precision(mAP_50:95). The results show that the improved YOLOv8s outperforms the algorithms of YOLOv5, TPH-YOLO, YOLOv7, and Swin transformer in the PI staging task, with an mean average precision (mAP₅₀₎ of 92.0%, and precision(P) of 86.7%, which are significantly better than those of the other models; moreover, compared with the original YOLOv8s, the improved YOLOv8s algorithm precision increased by 6.5%, mAP₅₀ and mAP_50:95 increased by 14% and 14.9%, respectively, and the recognition accuracy in each stage of PI (stage 1-4) was 89.3%, 84.3%, 73.2% and 100%, respectively. These results indicate that the improved YOLOv8s in this study can effectively recognize PI with different stages and provide an objective and reliable auxiliary tool for clinical diagnosis.(ChiCTR:250,289).