Research And Implementation Of Assisted Decision-Making Model For Diagnosis And Treatment Of Human Sports Trauma

Sports injuries refer to injuries that occur during human movement,whether in the life of the general public or in the sports career of professional athletes,their harm is universal and serious.The treatment of sports trauma mainly includes four parts: prehospital treatment,injury diagnosis,clinical treatment and follow-up rehabilitation.The most critical part of pre-hospital treatment is the timely treatment,and the early acquisition of accurate information of traumatic accidents has a significant impact on the treatment effect.In the injury diagnosis process,accurate diagnosis of critical injuries is the key point to improve the prognosis.In the clinical treatment segment,accurate pharmacological intervention for personalized information of the injured person and specific complex symptoms is the difficult point.With the development of artificial intelligence,computer-aided diagnosis and treatment has achieved results and shown good prospects in many disease areas,but it is still underappreciated in sports trauma research.In this thesis,the key aspects of assisted diagnosis and treatment for sports injury are studied,pre-hospital body collision part identification,assisted CT diagnosis of head injury and medication recommendation.And the sports injury assisted diagnosis and treatment platform is designed and implemented.The main work of this thesis is as follows.1.To address the problem of lack of accurate body part identification in injury detection,we proposed the body collision part identification based on Skeleton and Heatmap Fused Model(SHFM).The test results show that the collision recognition rate of SHFM exceeds 97.01% for the head and hip,which are the most critical parts of sports injury,and the accuracy of collision recognition for all parts reaches 98.66%.2.To cope with the significant morphological differences of brain hemorrhage lesions in head CT diagnosis,Low-level Context Attention Module(LCAM)is proposed,and the brain hemorrhage detection model LCAM-Yolo is constructed based on the object detection framework Yolo.LCAM enhances the model’s attention to low-level features and incorporates Coordinate Attention(CA)mechanism for contextual information fusion,which ultimately improves the network’s ability to detect brain hemorrhage.The experimental results show that LCAM-Yolo achieves 94.11% m AP in the BHX dataset.3.To handle the problem of missing mutual information modeling between multiple visits,the Self-Attentioned Medication Recommend Network(SAMRN)based on the self-attention mechanism is proposed.The SAMRN uses an encoder-decoder structure for feature extraction,and achieves an F1 value of 68.08% and a PRAUC value of 77.38%on the Mimic dataset.4.Based on SSM,Element-ui,vue.js and other frameworks,we designed and implemented a sports injury assisted diagnosis and treatment platform with "browserserver" as the architecture.The platform has the functions of user management,perception of injury accident,diagnosis assistance,treatment assistance,patient management and so on.