Automated Bone Age Assessment Based On Deep Convolutional Neural Network

Bone age assessment is the best way to judge the growth and development of adolescents and children,which has been widely used in clinical medicine,sports and judicial judgment.Because the wrist bone can reflect the overall bone growth and development more accurately,and has the characteristics of easy shooting and small radiation dose,the development of wrist bone is generally used as the evaluation standard of bone age at home and abroad.The traditional bone age assessment method is based on the degree of development of ossification centers in X-ray films of hand bones,and the artificial bone age is inferred by doctors.It has strong subjectivity,large random error,complex identification process and long recognition period and so on.In order to overcome the shortcomings of traditional bone age assessment methods,we proposed an automatic bone age assessment method based on deep learning and convolution neural network.The research contents and achievements are indicated as below:(1)To research the advantages and disadvantages of existing bone age assessment algorithms and methods,and to provide ideas and inspiration for the research and implementation of deep convolution neural network algorithm.At the same time,to research the latest research results of deep learning at home and abroad,taking into account the structural characteristics and advantages of existing neural network models,and to explore the applicability,stability and reliability of convolutional neural networks with different structures in the field of bone age assessment.(2)The preprocessing of hand bone X-ray image data is studied.In order to remove the background information(scale,artifact,noise,etc.)outside the main body of hand bone X-ray image and solve the problems of uneven gray distribution and non-alignment of spatial coordinates in the image,the U-Net model for biomedical image segmentation is firstly used to segment hand bone X-ray image.Then,histogram equalization is used to make the images with obvious gray differences have similar contrast.Finally,the key point detection model based on VGG module is used to realize image registration by affine transformation,and the hand bone X-ray images with separation of subject and background,contrast equalization and spatial coordinate alignment are obtained.(3)The automatic feature extraction of hand bone X-ray image based on deep learning and convolution neural network model is studied.The regression model of automatic bone age recognition based on DenseNet and the classification model of automatic bone age recognition are designed and implemented respectively.The performance and effect of automatic bone age recognition are verified by grouping comparative experiments.(4)An automatic bone age assessment system based on deep convolution neural network is designed and implemented.The deployment method of deep learning model is studied.The trained bone age assessment model is deployed online by using TensorFlow Serving high performance open source library launched by Google,which provides stable,efficient and convenient diagnostic assistance and decision support services for doctors and users.