Abstract:
The goal of this thesis is to study the use of neural networks for radiological bone age assessment from hand and wrist x-ray images is done. Carpal bones have been considered for bone age assessment. While the both semi-automatically and manually marked carpal bone features are given to our system as inputs, bone age is produced as an output. Additionally, chronological age, radiologist readings and sex information are used besides carpal bones and nally, the results are investigated. In this study, real data sets have been used. This study is important because a very simple and e cient method by using all 7 carpal bones is developed for assessing the bone age of children instead of the complicated methods in the literature. This semi-automated method also improves the time e ciency compared to the widely used manual methods such as GP, TW2. Inclusion of carpal bones for assessing bone age of children is mandatory. However, due to various factors including the uncertain number of bones appearing, non-uniformity of soft tissue, low contrast between the bony structure and soft tissue, automatic segmentation and identi cation of carpal bone boundaries is a hard endeavor. In this study, semi-automated carpal bone segmentation and age assessment software is developed and implemented. Also, neural network classi cation is used to assess the bone age depending on the selected features from carpal bones. In our application, 236 training images and 58 test images are used for 0 to 7 age group. After application, it is illustrated that results are considerably comparable with both chronological bone age and the two radiologist readings. We therefore conclude that the developed system may replace the manual methods for improved speed and comparable accuracy.|Keywords: Computerized bone age assessment, Greulich and Pyle (GP) method, Tanner and Whitehouse method, neural network, carpal bone
