A New Pediatric Bone Age Assessment Using Cluster based Lightweight U-Net Architecture Multi-Scale Convolutional Network

Abstract

For diagnosis, major indicator is age of bone and it is used to find timing of various diseases. Growth and development are computed using bone age assessment, in addition to that used to diagnose and treat childhood diseases mainly with the use of X-ray images. Also it considered as the contemporary standard clinical practice for diagnosing metabolic and endocrine disorders in childhood development. The outcome says that every single age can be estimated by looking at the bones of the child if the infant has this skeleton and what old the baby is. The main reason for this BAA is to know how the bone grows when the child develops. Segmentation of the hand bone is mainly needed at this point to explain the properties of the hand bone in detail in medical records with x-ray images. The newly developed lightweight multi-scale U-Net convolutional neural network is used with x-ray images in a previous study for the betterment analysis of BAA. Today, however, this approach is limited to segmentation maps in pixel forms and it has been addressed by developing a new model based on profound education that considers the field of interest within and outside the field and the dimensions of the borders during the training using K-mean clustering algorithms as a pre-segmentation constraint. U-Net architecture forms the base for design of lightweight hand bone of children and better results of segmentation is achieved by this, especially for BAA segmentation of hand’s small bones. For both learning and analysis this method requires images of the entire hands and various parts of the body. This method allows to measure the significance of the automatic examination of the age of the bone of each hand of x-ray images. The proposed approach allows for measurement of age of bone using other common methods while further evaluating the quality of the proposed method during developmental stages.