Reliable Diagnosis Of Synovitis Based On Musculoskeletal Ultrasound Images

Metacarpophalangeal(MCP)arthritis is a kind of rheumatoid arthritis,with the characteristics of onset hidden and slow progression of symptoms.The main clinical manifestation of the early stage is synovitis.If the disease is not diagnosed and treated in time,as the disease progresses the patient’s bones will be affected and the clinical symptoms will also change from mild swelling and pain to joint disease and the loss of joint function.In serious cases,it may even cause disability.Because of its advantages of noninvasive,low cost and fast imaging,ultrasonic detection has already become the main means of early clinical diagnosis of MCP synovitis.Currently,relevant reading and analysis work are still completed by professional doctors manually which may inevitably introduce subjectivity to diagnosis for the reason of poor consistency among different doctors.Moreover,heavy reading and analysis work are time-consuming and will increase the workload of doctors to a large extent.This paper is devoted to the realization of intelligent grading of metacarpophalangeal synovitis,which is essentially an automatic classification method based on ultrasound images.Medical images datasets generally have the problem of small size,which fall far short of the demand of deep learning.Although using the traditional image augmentation methods can expand the dataset to a certain extent,the image diversity is not guaranteed,which may easily lead to the overfitting problem of the training model.In this research,we propose an automatic grading method for MCP synovitis based on augmented ultrasound images generated by generative adversarial network,which can not only expand the dataset,but also ensure the diversity of images,so as to avoid the overfitting problem.In the proposed method,we built a large-scale input HRGAN model for the ultrasound images of MCP synovitis and made a training dataset for the model by using traditional image augmentation method——multiple cropping.Then we trained the HRGAN model on the training set to get a generator that possessed the ability of generating fake images which could be confused with the real ones,then we used the well-trained generator to generate a large number of false images to professional doctors for screening and grading.The final classification dataset was composed of the false images returned by doctors and the original images,and the training of classification convolution neural network was done on it to realize the intelligent grading of MCP synovitis.The intelligent grading method proposed in this paper can automatically classify the ultrasound images of MCP synovitis.On the one hand it can provide learning materials for young doctors who lack relevant experience and assist clinical diagnosis to reduce the workload of doctors,on the other hand it can also provide standard reference for related clinical researches,especially for multi-center involved ones.At the same time,this method has a certain versatility for realizing intelligent diagnosis of synovitis in wrist joint,knee joint and other parts,which also faces the problem of small dataset size.In the future work,the method can be improved and has the potential to apply to all joints of the whole body.