Research On Bone Age Assessment Method Of Residual Network Combined With Efficient Channel Attention Module

Bone age assessment is a technique used in clinical diagnosis and medical treatment to determine whether there is a difference between a child's skeletal age and his or her actual age.The traditional bone age estimation method is based on the development of bone ossification centers in X-ray images of patients' hand bones,which is subjective,error,complex and time-consuming.In view of the shortcomings of current bone age assessment methods,an automatic bone age assessment method based on deep learning and deep convolution neural network is proposed.The main contents and research results are as follows:(1)A method of hand bone segmentation based on U-NET was proposed.The key to the study of hand bone segmentation is to remove as much useless noise as possible and preserve the relevant features of bone age assessment.This thesis analyzes the current research results of image segmentation and related methods,focusing on the characteristics of medical images and the particularity of image segmentation.In view of its particularity,we replace the original encoder of the U-Net model with the pre-trained VGG16 network on the Image Net dataset,and construct an improved U-Net-based hand bone segmentation model to achieve the goal of improving the quality of data image.(2)A residual network method combined with efficient channel attention for bone age assessment was proposed.The key to bone age assessment is to extract high-level semantic features associated with bone age from the images.Based on the functional requirements of bone age assessment and the characteristics of experimental data,the residual network is improved.Through introducing efficient channel attention model and stochastic depth algorithm,the relative loss function is designed,and a new model of bone age assessment based on residual network is constructed in order to restrain the problem of overfitting caused by small data size and alleviate the effect of uneven sample distribution on the model assessment performance.(3)To verify the effectiveness of the method of hand bone segmentation and bone age assessment.Analyze the real process of bone age assessment,simulate the situation of bone age assessment,and use the improved U-Net hand bone segmentation model and the residual bone age assessment model combined with efficient channel attention to conduct hand bone segmentation and bone age assessment.Analyze the experimental results to verify the effectiveness of the improved U-Net hand bone segmentation model and residual bone age network assessment model.(4)The hand bone segmentation module and the bone age assessment module in the automatic bone age assessment system are designed and implemented.In the design of sub-module,the modeling method is introduced in detail,including the case diagram,sequence diagram,class diagram,database diagram and the main code realization of the core function module.It is also ready to use Google's Tensor Flow Serving open source library to deploy a trained skeletal age assessment model online,so that more doctors and users can get accurate,efficient and convenient auxiliary diagnostic services.