Research On Bone Age Assessment Method Based On Deep Learning

Bone age assessment is an important clinical method to study endocrine problems and metabolic disorders in adolescents and children.Bone age assessment has been at the heart of many disease assessments since it was proposed in 1955.However,the method of bone age assessment has not changed much over the years,and most hospitals still use artificial comparison map to evaluate bone age.Such an approach has huge workload,serious resource consumption,and is easily affected by human factors,which makes the accuracy rate of assessment unstable.In view of the problems in the bone age assessment method,this paper,based on the deep learning method,studied the pre-processing method of bone age X-ray image and the training details of the convolutional neural network model,and completed the main research work and results as follows:(1)a preprocessing method of X-ray image of hand bone is proposed.According to the characteristics of the X-ray image of the hand bone,the image was processed before training the deep learning model.The gray histogram equalization method is used to enhance the contrast of the original image and increase the feature significance.Make target detection training set to train SSD model and detect several key positions in the image.By using the proposed algorithm,the posture of the hand is unified.Threshold segmentation combined with morphological algorithm was used to remove background interference and to segment the hand bone in the image.The proposed algorithm is used to transform the image into uniform size while maintaining the original scale.In this way,the training set of hand bone images removing the interference factors is obtained.(2)the training of the convolution neural network model is studied.The structure of the convolutional neural network was analyzed,the classical convolutional neural network model was compared and discussed,and the structure of the bone age assessment model was designed according to the requirements.The details of training,including data enhancement,hyperparameter adjustment and prevention of overfitting,were studied.A visualization method is designed for model visualization.(3)to test the effect of the convolutional neural network model completed by training.The error distribution diagram and model attention heat map are drawn and the model reliability is analyzed.A bone age measurement system was developed.The method is based on deep learning.After image preprocessing by multiple methods,bone age is evaluated by using convolutional neural network model,and good results are obtained.