Publication Date


Document Type


Committee Members

Dana L. Duren (Committee Member), Thomas N. Hangartner (Advisor), Julie A. Skipper (Committee Member)

Degree Name

Master of Science in Engineering (MSEgr)


Children's skeletons mature at different rates, and they can be affected by a variety of factors including disease, hormone imbalance or genetics. The assessment of skeletal maturity is a frequently performed procedure that allows the detection of hormonal, growth or genetic disorders. Several methods have been developed to estimate skeletal maturity. Most methods evaluate hand/wrist radiographs using indicators such as the ratios of various bone widths, the onset of the ossification of epiphysis and epiphyseal-diaphyseal fusion. Among those methods, the FELS method differs from others in the application of different grades to each indicator and the provision of a confidence limit of the determined skeletal maturity.

However, skeletal age assessment based on the FELS method, as with any method, is associated with observer variability. There is also increased pressure on pediatric radiologists to read more and larger sets of radiographs. These problems could be solved by an automated computerized method, which has the potential to reduce the time required to examine the image and to increase the reliability of the analysis. The aim of this project is the development of an automated computer-based analysis method to estimate skeletal age from hand/wrist radiographic images. Such images were obtained through not only traditional x-ray procedures but also from a dual x-ray absorptiometry scanner. The analysis was performed in several stages: the preprocessing step, the ROI extraction step and the indicator analysis step. The results obtained from the analysis were then integrated and used to calculate the skeletal age and its associated standard error. In this study, 174 left hand/wrist radiographs of children between the ages of 8 and 18 years were selected from the FELS Longitudinal Study; 100 of them were used for training and the remainder for testing. DXA images of the participants in the testing set were used to evaluate the possibility of assessing skeletal age based on DXA. The automated analysis was successful in approximately 90% of the training set, 85% of the testing set and 100% of the DXA image set. Manual intervention of the ROIs localization allowed the remaining images to be analyzed.

The grades of all the indicators together with the skeletal age of each participant generated from our analysis method were compared with the reference values provided by two well trained specialists at the Lifespan Health Research Center. Most of the indicators (85%) do not show statistical differences between the observation values obtained from our program and the reference values. By comparing the skeletal age estimated by our program and by the specialists, it was found that the analysis of the traditional x-ray images was fairly good; only 13.3% of the training set and 20.6% of the testing set show differences that are larger than one year. However, the results of the DXA images were worse; about 40.5% of this data set show a difference larger than one year. In addition, the indicators that could not be graded by our program do have some effect on the skeletal age assessment of the children from 8 to 18 years old.

Page Count


Department or Program

Department of Biomedical, Industrial & Human Factors Engineering

Year Degree Awarded