Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Orthopedic and Rehabilitation Engineering

Orthopedic and Rehabilitation Engineering - Poster Session A

Poster K5 - A novel approach to enhancing efficiency in biomedical engineering research prior to access to human subject data

Thursday, October 24, 2024

10:00 AM - 11:00 AM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

RJ

Ryan Jaworski

Student
Fairfield University
Carlstadt, New Jersey, United States

Co-Author(s)

Luke Bornstein

Undergraduate Researcher
Fairfield University, United States
Zuzanna Janusz

Undergraduate Researcher
Fairfield University, United States
DP

Donna M. Pacicca, MD

Pediatric Orthopedic Surgeon
Connecticut Children's Hospital
Hartford, Connecticut, United States
JD

John F. Drazan, PhD

Assistant Professor of Biomedical Engineering
Fairfield University, United States

Introduction: Biomedical engineering is the development of solutions to problems in human health. This often requires access to patient health information (PHI) such as clinical imaging, which requires significant regulatory oversight and permissions. Clinical imaging, such as Mechanical Axis Films (MAFs), can be an important part of an experiment, however, can only be accessed once permission is applied for and received, often taking significant time and inhibiting the progress of research. The purpose of our project was to create a new paradigm for preparing for imaging studies through the creation of a synthetic data set, which mimics the properties of a clinical data set in size, structure, and relevant measurements, while avoiding the use of human subject data. Our approach has potential to allow for more efficient research projects by allowing significant technical and methodological progress to occur prior to access to patient data.

Our original project, developed in collaboration with Connecticut Children’s Hospital, was to analyze the repeatability and reliability of Mechanical Axis Films to identify and track the development of Limb Length Discrepancies in pediatric populations. Prior to the start of the project with human subject data, we needed to do the following: determine a method to randomize the MAFs while blinding the image raters, develop a workflow for analyzing the images, and create a program to derandomize the data and prepare it for statistical analysis. While waiting for IRB approval, we created a synthetic data set to model MAFs while accurately simulating a pediatric patient population with LLDs.

Materials and

Methods: The synthetic data set consisted of 30 images; 10 patients who have each had three MAFs done. MAFs were mimicked by taking pictures of wooden planks to model the femur and tibia with a four-foot ruler placed between the two “legs” for calibration and measurement. Lengths of the wooden planks were determined using online CDC growth charts to estimate bone length of the femur and tibia. Of the 10 patients in the synthetic data set, four had a structural LLD in which a femur on one side of the body was shorter than the other, three had a structural LLD in which a tibia on one side of the body was shorter than the other, and three patients had hemihypertrophy, in which both the femur and tibia on one side of the body were shorter than the other. Seven of the synthetic patients had clinically relevant LLDs while the remaining three had clinically irrelevant LLDs ( < 2cm).

We generated a MATLAB code to randomize the images into six different blinded image data sets, each with their own linking key (3 raters, 2 repeats). We decided to make measurements in ImageJ of the locations of the proximal and distal ends of the tibia and femur and recorded in an Excel file. We then created an additional code to derandomize the data. To assess reliability and repeatability of limb length measurements, we used ICC testing for inter- and intra-reliability.

Results, Conclusions, and Discussions: Across both days of measurements and the three raters, there is varied error ranging from relatively low (0.51cm) to high (1.70cm). Intra class correlation coefficient type 3,k statistics were performed. ICC statistics were found to be 0.3253 with a [-0.2369 0.6565] 95% confidence interval for day 1 and 0.1855 with a [-0.4933 0.5853] 95% confidence interval for day 2 between each of the three raters. These scores indicate poor reliability of the reliability and reproducibility of measurements taken on the synthetic data set.

The creation of a synthetic data set that mimics the properties of clinical imaging data has allowed us to hasten the research process in biomedical engineering. By developing a model that accurately simulates MAFs of pediatric patients with LLDs, we have enabled substantial technical and methodological advancements to be made prior to obtaining patient data and IRB approval. Our synthetic data set approach allowed us to implement and test our entire workflow, from randomization and blinding to image analysis and statistical evaluation. The ICC values indicated poor reliability of measurement across novice raters, however, it has illustrated the importance of developing better training and documentation prior to the start of our study. Although the low error in UG2’s measurements indicated that a low error can be achieved, the ICC values illustrated that a standardization of measurement using ImageJ is necessary before the analysis on clinical imaging. The development and use of a synthetic data set provided a valuable opportunity to refine our methodologies and address potential issues in a controlled, low stakes, environment.

Acknowledgements (Optional):