Research And System Implementation Of Abnormal Hand Bone Detection Based On CNN

Medical images play an important part in the disease diagnosis course by doctors.With the continuous development of society,the means of imaging are increasing gradually,and the number of images is also increasing with the rapid growth of the population.A lot of images not only put a lot of pressure on doctors' work,but also lead to the increasing probability of misdiagnosis and missed diagnosis caused by overwork of doctors.Based on these conditions,computer aided diagnosis becomes one of the research directions of the current medical image,it can carry on the pretreatment and classification of medical images.With the development of computer hardware technology,computers are far superior to humans in recognition,the ability of detail feature collection is strong,it has positive significance to reduce the error probability of leak detection inspection and to relieve the pressure of the doctors.At present,we usually use the current relatively popular convolutional neural networks to deal with medical images.For medical image,convolutional neural networks perform well in the process,many researchers apply it to the medical image diagnosis.due to some features of the disease are very small,so convolutional neural networks need more convolution layers to study these characteristics.But it is difficult to receive a large number of manually labeled medical image data sets.However.if convolution layers are more,more parameters will be introduced and we need so many cases to train them,so a small amount of training data set is difficult to train many parameters,even existed overfitting problem.This paper is mainly to study the hand bone disease,Hand bone X-ray images are from MURA data.The hands bone diseases are mainly divided into bone injury and joint disease.It brings great trouble to people's life for a long time,If the disease is not treated in time.Then it will cause impact,so in this paper,the experiments are going to use the convolutional neural networks based on transfer learning to classify the hand bone X-ray images,in comparison with ordinary neural networks,it proves the effectiveness of transfer study,in order to avoid the abnormal situation in diagnosis,improve the classification precision of computer,HOG+ SVM classification experiment was carried out.Based on the two classification methods,the hand bone X-ray classification system was integrated.It presupposes that abnormal images are first diagnosed and processe.The X-ray image was preprocessed can reduce the work pressure of medical staff,and the system had a good classification to the examples.