Ahmad Shirzadi

Determining bone age is one of the radiological diagnostic methods that is used in the daily work of most physicians and helps physicians in calculating the maturity of the skeletal system and the growth rate of children’s skeletons. Measuring bone age is a good criterion for diagnosing precocious puberty as well as underlying diseases, such as hypothyroidism and pituitary diseases that affect the growth of bone centers and can be detected from bone. Bone age should be commensurate with the calendar age. This indicates the normal growth of the skeleton in proportion to the age of the person. In determining bone age, if bone age is in accordance with the calendar age, bone growth is normal, but if it is less than the calendar age, it is called bone age delay. Different systems have been proposed to estimate the age of skeletal bone, which more or less have shortcomings and problems. In this dissertation, the aim is to present a system that estimates the age of a person’s bone from radiological images of the hand using deep neural networks. Since the hand radiographs here are labeled with age, the bone age of the people we estimated should have the least error relative to the given age. Our architecture has a supervised learning method and the neural network used is the feed neural network. Activation functions such as Relu and Softmax are used, and the number of layers, 24, and the architectures used include 2D Convolution, Batchnormalization, MaxPooling, Flatten, and Dence. The architecture is compared with VGG16, DenceNet and Inception methods and we see that it has higher accuracy and efficiency compared to other existing methods