Bone age detection with deep learning methods

Determination of bone age is an important procedure for diagnosing various diseases, monitoring bone development or understanding the presence of hormonal problems. This determination is especially important in childhood, during the fastest stages of growth. In cases such as an excessively long or excessively short neck, it is possible to prove the individual's childhood or adulthood; it is also possible to determine the last point where the bones can develop until the end of puberty, i.e. the length of the height, with bone age assessment. Each individual has two ages: bone age and chronological age. Bone age is the degree of skeletal maturity of individuals. Chronological age, on the other hand, is the age we know, which is obtained by calculating the years from the date of birth to the current date; it is the age we know when we answer the question of how old we are. The main purpose of bone age determination is to determine whether there is a problem in the development of the individual by evaluating the difference between these two ages. In this study, x-ray images of 12611 left hands of individuals between the ages of 0-18 were compared with Convolutional Neural Network (CNN) and the Inception V3 model, which gave the best results, was improved and a new model was created. VGG-16, Inception V3 and MobileNet architectures were used in the experimental studies to compare the new model and all architectures were compared according to Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Correlation Coefficient (CC) values. The new model gives much faster results than the classic Inception V3 and is also more accurate. When compared with the MAE values of other studies in the literature using the same dataset, it is observed that the model gives good results.