Dana L. Duren (Committee Member), Thomas N. Hangartner (Committee Member), Julie A. Skipper (Advisor)
Master of Science in Engineering (MSEgr)
To understand the roles of various genes that influence skeletal bone accumulation and loss, accurate measurement of bone mineralization is needed. However, it is a challenging task to accurately assess bone growth over a person's lifetime. Traditionally, manual analysis of hand radiographs has been used to quantify bone growth, but these measurements are tedious and may be impractical for a large-scale growth study. The aim of this project was to develop a tool to automate the measurement of metacarpal cortical bone thickness in standard hand-wrist radiographs of humans aged 3 months to 70+ years that would be more accurate, precise and efficient than manual radiograph analysis.
The task was divided into two parts: development of automatic analysis software and the implementation of the routines in a Graphical User Interface (GUI). The automatic analysis was to ideally execute without user intervention, but we anticipated that not all images would be successfully analyzed. The GUI, therefore, provides the interface for the user to execute the program, review results of the automated routines, make semi-automated and manual corrections, view the quantitative results and growth trend of the participant and save the results of all analyses.
The project objectives were attained. Of a test set of about 350 images from participants in a large research study, automatic analysis was successful in approximately 75% of the reasonable quality images and manual intervention allowed the remaining 25% of these images to be successfully analyzed. For images of poorer quality, including many that the Lifespan Health Research Center (LHRC) clients would not expect to be analyzed successfully, the inputs provided by the user allowed approximately 80% to be analyzed, but the remaining 20% could not be analyzed with the software.
The developed software tool provides results that are more accurate and precise than those from manual analyses. Measurement accuracy, as assessed by phantom measurements, was approximately 0.5% and interobserver and intraobserver agreement were 92.1% and 96.7%, respectively. Interobserver and intraobserver correlation values for automated analysis were 0.9674 and 0.9929, respectively, versus 0.7000 and 0.7820 for manual analysis. The automated analysis process is also approximately 87.5% more efficient than manual image analysis and automatically generates an output file containing over 160 variables of interest. The software is currently being used successfully to analyze over 17,000 images in a study of human bone growth.
Department or Program
Department of Biomedical, Industrial & Human Factors Engineering
Year Degree Awarded
Copyright 2008, all rights reserved. This open access ETD is published by Wright State University and OhioLINK.