The Method And System Of Hamate Maturity Assessment Based On Convolution Neural Network

Bone age is a reliable indicator for evaluating growth and development.The accuracy of the evaluation results is of great significance for understanding the real-time growth and development of adolescents and children.Accurate evaluation of the maturity of the reference bone of the wrist is the key to discerning the bone age.As one of the most weighted reference bones in the Chinese standard of wrist bone development,the hamate has a crucial role in accurately assessing the bone age.However,in the traditional manual evaluation method,there are few professional doctors and it takes a long time to read the film,which makes it difficult for the manual evaluation method to break through the bottleneck of large workload and low efficiency.Therefore,it is of great social significance and practical value to replace traditional manual evaluation with computer evaluation methods.The main difference between different maturity levels of the hamate in the wrist X-ray image is the difference in the texture and morphological features of the bone mass,while the feature differences between adjacent or similar levels are smaller;when the maturity level is higher,the hamate overlaps with many bones on the X-ray image.These factors bring more challenges for the computer to accurately evaluate the maturity level of the reference bone;and in the research for single reference bone at home and abroad,there are widespread problems such as broad division of mature stages and missing position detection.In view of the above problems and difficulties,this paper uses the physical fitness test data of adolescents and children collected in the "Campus Charity Walk" project in Zhejiang Province to conduct related research on the intelligent evaluation of hamate.The main work is as follows:(1)Aiming at the complicated situation of the wrist,an adaptive extraction method of the hamstring feature region of the wrist was proposed.According to the height and age information of the person to be detected,the method can personally generate a characteristic area of the hamate that matches its own growth condition,and then effectively reduce the interference caused by the surrounding bone mass and muscle tissue in the grade evaluation.The effects of different feature region extraction methods on the accuracy of maturity level evaluation of convolutional neural networks were compared in experiments,and the feasibility of the method was verified.(2)Aiming at the feature with small differences in features between maturity levels of hamate images,this paper proposes a feature-enhanced residual network model,which enhances the feature extraction capability of the network through a widened residual block and a two-pass parallel approach,and An improved loss function is used to reduce the impact of cross-level errors on the accuracy of bone age assessment.The experimental results show that the accuracy of the network model in the evaluation of hamate maturity level is 89.4%,and there are no errors that span more than two levels.(3)Due to the differences in the structural design of different convolutional neural networks,the extracted feature information is also different.In order to further improve the accuracy of hamate maturity assessment,this paper proposes a joint probability judgment algorithm based on multiple convolutional neural networks.This algorithm can use the outputs of multiple convolutional neural networks to make joint probability judgments to obtain the final result.Hamate maturity level.The experimental results show that the accuracy of the hamate maturity level evaluation can be further improved to 90.4% by this algorithm,and no errors over two levels have appeared.(4)Design and implement a web-based hamate maturity level evaluation system.The adaptive extraction model of crocodile feature region,feature-enhanced residual network model,and probabilistic joint algorithm proposed in this paper are integrated into the system to help experts quickly evaluate the maturity level of crocodile.At the same time,this system can further expand the data set and provide a data basis for subsequent research.