The evaluation of bone development is a complex task since it may cause intra-observer and inter-observer differences, and it may also vary depending on the variations observed very often in the normal children. In this thesis study, a computer-based diagnostic system to detect the bone age of the children aged between 0-6 years was proposed. In the first phase of the study, primarily the image processing procedure was applied on the x-ray images of the left hand-wrist of children aged between 0 and 6 from different ethnic groups and totally 9 features corresponding to the carpal bones and distal epiphysis of the radius bone along with some physiological attributes of the children were obtained. Afterwards, with the help of gain ratio, the best 6 features were selected for the classification process. These selected features were classified by using different artificial intelligence techniques such as C4.5, simple Bayes, k nearest neighbor algorithms and support vector machines. In the second stage of this thesis study, three different training data reduction methods were investigated so as to produce faster results in terms of execution time of the support vector machines at the stage of system training that is carried out by support vector machines to detect the bone age. In the third and final phase, a new particle swarm optimization based training algorithm was suggested for the bone age assessment system based on support vector machines in the children aged 0-6. By means of the particle swarm optimization algorithm, a new sample in each class was created to represent this class best, which is used for training. Thus, it provided the support vector machines with the training by the new samples, without depending on the currently used samples that are used for system training. In this proposed new method, system training is carried out by accepting the new the samples as the new support vectors, which were obtained by particle swarm optimization algorithm. |