Segmentation And Feature Extraction Of The Key Bone Structures In Automatic Bone Age Assessment

Bone age assessment is an important way to evaluate the degree of one's physical development.It is widely used in kinematics,forensic medicine,pediatrics and other fields.Traditional bone age assessments rely on experts to read a person's left hand X-ray image,which is subjective and time-consuming.With the rapid development of the digital medicine,the demand for computer-aided automatic bone age assessment is also increasing.There are two difficulties restricting the development of automatic bone age assessment.One is that during the development the shape of the wrist bones is always changing and it tends to fuse at the late period of development,which makes it difficult to segment.The other is how to interpret the bone age assessment criteria described in natural language into quantifiable image features.This thesis mainly studies the segmentation and feature extraction technology of the key bone structures in automatic bone age assessment.The main works are listed as follows.In this thesis,the key bone structures in automatic bone age assessment are defined and a bone structure localization method is proposed based on the physiological structure and gray level characteristics of left hand.Several common segmentaion technologies are studied,and a multi-stages-multilevels CLM Segmentation algorithm is proposed,which takes the different stages of bones development into account,establishes a multi-stage shape model and uses multilayer search strategy to ensure the correctness of the model convergence.The thesis studies the small sample size high dimension(SSHD)problem in point distribution model based segmentaion methods and proposes solutions from two ascpects.During appearance model generation,we extend the sample size by introducing gray level transforms and acceptable sample error.During statistical shape model generating,we increase the generalization ability of the model by relaxtion the correlations of certain landmarksThe thesis designs and extracts the geometric and texture features of the key bone strucutres based on the clinical bone age assessment standard and experts' opinions,and then gives the feature selection suggestions through the classification experiments.