Design And Implementation Of A Wound Care Assistant System Based On Deep Learning

The recovery period for chronic wounds usually lasts for several months,and it is easy to become infected during the recovery process.During the recovery process,professional nurses often need to care for the wound face-to-face with the patient.However,high cost and existing medical resources do not support long-term hospitalization of patients,so an auxiliary wound care system needs to be developed.This paper designs and develops a chronic wound care assistant system based on deep learning,which assists patients in wound care,reduces the number of patients traveling to and from the hospital,and reduces the burden on doctors.In this paper,the chronic wound nursing assistant system realizes the functions of user login,user management,information input,information query,wound image segmentation and recognition.The wound image segmentation and recognition function is the core function of this system.In view of the fact that there are few large-scale public wound image segmentation data sets at present,the advantages of deep neural networks cannot be exerted.The existing research is based on small data sets for training,and the generalization of the model cannot be guaranteed.This paper proposes to use transfer learning to perform wound Image segmentation.First,the Res Net50 network trained on the large public dataset Image Net is used as the feature extractor,and then the Danet network is used for retraining with the small wound image segmentation dataset produced.Experiments show that this method can accurately segment wound images.Compared with the model,this model achieves state-of-the-art results on the MIOU metric.Since multiple wounds at different recovery stages may appear in the wound images captured by the user,wound image recognition cannot simply classify the images.This paper proposes to use the improved YOLOX algorithm to identify wound images.The four stages of wound recovery（dry necrosis stage,verification reaction stage,granulation growth stage,and epithelialization stage）are used as four different targets,and the improved YOLOX is used for identification.The FPN feature enhancer of the original YOLOX only uses the three high-level features extracted by the CSPDarknet network for fusion.This paper believes that the wounds in the four different stages of the recovery process are quite different in shape and color,so the underlying layer extracted by the CSPDarknet network is common.The features should also be fused by the FPN feature enhancer.In this paper,the low-level features and high-level features extracted by the CSPDarknet network are fused with the FPN feature enhancer and then passed into YOLOhead for classification and prediction.Experiments show that the improved YOLOX in this paper can accurately identify the recovery stage of the wound on the wound image taken by the user.Compared with the original YOLOX,the mAP,mAP₅₀,and mAP₇₅are improved.