Research On Upper Limb Fracture Diagnosis Technology Based On Convolution Neural Network

There are more than 1.7 billion people who are tired of the diseases of bones around the world.As a dominant diagnosis measure of bone diseases,whose performance are depending on the practical experience of radiologists seriously with a certain degree of uncertainty.The total number of radiologists can’t meet the increasing demands of skeletal disease nowadays.The automatic recognition and classification method of X-Ray image data based on artificial intelligence technology is expected to partially replace doctors’ manual diagnosis and effectively alleviate the contradiction between the needs of doctors and patients.In recent years,with the continuous development of deep learning,the application of convolutional neural network in medical image processing has made remarkable progress.This topic intends to apply convolution neural network to the classification and detection of upper limb bone fractures,and study the key algorithms and technologies such as automatic recognition of upper limb bone types and location of upper limb fractures.The main work of this study is as follows:(1)The limb bones dataset construction: The injury classification of the upper limb bones dataset and the position dataset is constructed and labeled.The image enhancement process is carried out by the adaptive contrast restriction algorithm,which makes the bone edge more obvious and the location of the focus more visible.(2)The upper limb bone classification model and upper limb bone position model is established: the IDNet convolution neural network based on Inception Net-v3 and Dense Net was proposed for the bones image feature extraction.After training with the upper limb bone injury model,we can get about 92% classification accuracy with IDnet,which is 5.4% higher than Dense Net and 8.8% higher than Inception Net-v3.We can get 99% classification accuracy with IDnet after training with the upper limb bone parts model.(3)The upper limb bone fracture detection model is discussed: The performance of the Faster-RCNN algorithm and YOLOv5 algorithm was compared in the classification of the fracture detection with the upper limb bone fracture detection dataset.A modified Faster-RCNN was proposed,which replace the VGG16 trunk net structure with Inception-Res Net-V2.After the experiment,the performance of the improved fast RCNN is the best,and the accuracy of the improved algorithm is improved by 8%.Therefore,as an algorithm of the upper limb bone fracture detection model,the final average accuracy is 88%.After testing,the technical research on the diagnosis of upper limb fracture has completed the diagnosis of upper limb fracture,which provides favorable help for the subsequent development of the upper limb bone diagnosis algorithm model.