Research On Key Technologies Of Bone Age Automatic Assessment And Its Prototype System Development

As a primary indicator for assessing the growth and development of youth,bone age has important value in medical diagnosis,athlete selection and judicial identification.At present,the main means of clinically assessing bone age is manual diagnosis.On the one hand,experienced doctors are scarce,on the other hand,subjective differences are large and take a long time.As a result,pediatricians are increasingly concerned with the development of computer-aided assessment of bone age technology.There are three main difficulties in the current automatic assessment of bone age: firstly,it is difficult to accurately segment the target bone structure due to the large morphological changes in the development of the hand bone;secondly,it is difficult to effectively parameterize the image features described in current bone age assessment criteria and extract valid bone shapes and texture features;Thirdly,it is difficult to find a link between skeletal maturity and bone image characteristics,so that the advantage of the machine's evaluation of bone age is not obvious compared with manual interpretation.In order to solve the above difficulties,this paper studies several key technologies in the automatic assessment of bone age.The main contents are as follows:(1)RUS bone accurate segmentation method based on AHSM.Aiming at the difficulty of bone segmentation,an adaptive hand bone segmentation model(AHSM)is proposed.According to the characteristics of hand bone development in different periods,multi-stage shape model and multi-level projection model can be established,which not only ensures the strong local constraint ability of the distribution model,but also improves the bone recognition rate and realizes accurate segmentation of the target bones adaptively.(2)RUS bone shape and texture feature extraction method.The shape and texture parameters of RUS bones were successfully extracted by statistical shape model method and Delaunay triangle mesh generation method.Compare the extracted key features of the bone and select image features with good bone age discrimination.(3)A bone age assessment method that combines the shape-texture dual features.Two algorithms for shape-texture fusion features,namely multiple linear regression and support vector regression,are used to establish the bone age regression model,not only prove the rationality of the obtained features,but also show high bone age estimation accuracy in both training sets and test sets.It meets the clinical requirements for the accuracy and stability of bone age.(4)Development of a bone age automatic assessment prototype system.According to the modular design of the system,an almost fully functional bone age automatic assessment prototype system(Auto Bone AGE Standalone)was developed.The system integrates key techniques such as hand bone adaptive segmentation,shape-texture feature extraction and bone age automatic assessment,and realizes the bone age automatic detection of 1?18-year-old hand bone image,which is recognized by imaging experts and pediatricians.