Bayesian estimation of growth age using shape and texture descriptors

This paper presents an automated growth estimation system based on Bayesian principle by using knowledge-based vision methods to localise and segment bones in hand radiographs. Traditional manual methods have been tedious and prone to inter and intra observer inconsistencies. A segmentation algorithm known as active models (ASM) followed by a hierarchical bone localisation scheme is used to detect bone contours and also to produce a shape descriptor of bone development. Traditional image processing techniques are applied to generate different descriptors for bone shapes. A Bayesian decision-making algorithm is then applied to the descriptors for growth estimation purposes. The estimation accuracy was 85% for females and 83% for males, which suggests that the proposed approach has a potential application in paediatric medicine.