Automatic Evaluation Of Bone Age Based On X-ray Images

Bone age assessment is a commonly used clinical method to study endocrine,genetic and growth disorders in adolescents and children.The bone age assessment is usually based on the identification of skeletal developmental characteristics to obtain a numerical evaluation of human development.At present,bone age assessment in China is mainly accomplished through manual reading of X-ray hand bone images by doctors.However,this method has a huge workload,consumes large amounts of resources,takes a long time to measure,and is susceptible to human factors.This paper based on the X-ray hand bone image uses computer vision,machine learning methods to study the X-ray hand bone image preprocessing and segmentation,feature extraction,automatic bone age assessment,lack of image samples and other core issues.The main research work and achievements are as follows:(1)Study the X-ray hand bone image preprocessing and segmentation techniques.X-ray hand bone images in our data set have different resolutions,noise pollution,background interference and uneven gray scale distribution.In order to solve these problems,this paper de-noises images and scales them to a uniform resolution size.Using histogram equalization to even out the distribution of gray levels between images.At the same time,several commonly used image segmentation algorithms are used to segment the image in order to split the hand bone area.A cropping method for the X-ray hand bone image is proposed to crop the image,so as to cut out the background as much as possible and preserve the hand bone region.(2)Study the solution of the small sample space of the X-ray hand bone image and compare using geometric strategy,GAN strategy,and elastic distortion strategy to generate samples.Experiments have shown that due to the fixed orientation and location of the hand bones in the image samples,it is not suitable to generate samples using geometric strategies.Constrained by the capacity of the original sample,the quality of the sample generated by the GAN strategy is poor and GAN strategy is equally unsuitable for generating samples.Only the elastic deformation strategy can generate better quality samples without destroying the spatial distribution invariance of the original sample data.(3)Study artificial extracting features of the hand bone image and combing with SVM for evaluating the bone age automatically.The SVM-based automatic bone age assessment method mainly adopts artificial features,extracts SIFT features,LBP features,and GCLM features from images,combines these features,and then these features are used to train SVM.The SVM-based bone age assessment method has certain bone age automatic assessment capability.(4)Study deep learning to automatically extract X-ray hand bone image features,using convolutional neural network for automatic bone age assessment.Based on the residual network and spatial transformer,a new ST-ResNet network model is proposed.Compared to the automatic bone age assessment method of artificial extraction features,the model shows better evaluation accuracy.