Research On The Application Of Imaging Diagnosis Of Ankle Fractures Based On Deep Learnin

Ankle fracture is currently the most common type of foot fracture injury,and if the best time for treatment is missed,it can easily lead to secondary cartilage injury,joint laxity,synovitis,and traumatic osteoarthritis and other related complications;therefore,it is especially important to diagnose the type of ankle fracture image in a timely and correct manner.In the current clinical diagnosis,X-ray examination has developed into a routine means of ankle fracture diagnosis,but because X-ray ankle fracture images are usually accompanied by a large amount of background noise interference,resulting in a difficult clinical diagnosis and a large workload.In order to reduce the clinical workload and improve the correct rate of ankle fracture image diagnosis,this paper carries out the research on intelligent diagnosis of ankle fracture images based on convolutional neural network,which mainly includes the following aspects:(1)The image preprocessing and data enhancement methods were studied to address the problem of the current small ankle fracture medical image dataset.First,a preprocessing method was designed for ankle fracture medical images,and the restricted contrast adaptive histogram equalization(Contrast Limited Adaptive Histogram Equalization,CLAHE)algorithm was used to perform preprocessing operations on ankle fracture X-ray images to achieve image enhancement.Secondly,data enhancement methods such as random flip,random crop,and random brightness adjustment were used to expand the sample size of the data set.(2)Two transfer learning methods based on the Res Net-50 network model were designed to address the problems of long training time and low classification accuracy of Res Net-50 network in medical image classification of ankle fractures.First,Res Net-50 was used as the base network for training and testing the data.Second,to improve the classification accuracy and shorten the training time,two transfer learning methods were designed: pre-trained model as feature extractor and model fine-tuning.The experimental results show that the highest accuracy improvement of ankle fracture image classification is achieved after optimization of Res Net-50 by model fine-tuning method,and the training time is also effectively reduced.(3)To further improve the classification accuracy of the transfer learning model for medical images of ankle fractures,a classification method based on the attention mechanism and Res Net-50 is proposed.First,a new attention mechanism is designed: Dual Attention Mechanism(DAM),which not only captures effective information across channels,but also senses the size and location of input features,enabling the model to extract useful features more accurately.Secondly,the DAM-Res Net classification network model is proposed by improving the Res Net-50 network model by embedding the DAM into the residual blocks,thus enabling the network to better extract features from medical images of ankle fractures and improve the classification accuracy of the algorithm.Finally,the comparison of experimental results verifies that the DAM-Res Net classification network model can effectively improve the classification accuracy of ankle fracture medical images compared with other methods.(4)The study designed an ankle fracture image diagnosis system.In the requirement analysis and overall design phase,the following functions were implemented considering the scenarios in which users use this system: after users upload ankle fracture images,the system automatically completes image pre-processing,then invokes the algorithm for ankle fracture classification prediction and visualizes the related results.In the testing phase,we tested and validated the system,which proved the effectiveness of the algorithm proposed in this paper.The system is easy to operate,and the classification prediction of ankle fracture images meets expectations and can be used in clinical practice.