Skip to main content
2025 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Biomedical Imaging and Instrumentation

Biomedical Imaging and Instrumentation - Poster Session E

T20 - Inter-Reader Variability in 3D VS 2D Cardiac MRIs

Saturday, October 11, 2025
10:00 AM - 11:00 AM PDT
Location: Exhibit Hall F, G & H

Introduction: : Left ventricular ejection fraction (LVEF) is a critical parameter in assessing cardiac function, commonly used to diagnose and monitor conditions such as heart failure. Cardiac MRI is widely used to measure LVEF because it offers detailed images and reliable measurements. The accuracy of LVEF from 2D vs 3D imaging has been previously reported, however human interpretations introduce inter-reader variability impacting the precision and consistency of clinical assessments. This variability may depend on the imaging methods used—particularly between two-dimensional and three-dimensional analysis. Our hypothesis is that because 2D imaging only captures a single slice, this will result in a higher inter-reader variability. In comparison, 3D imaging considers all the slices which in theory would average out any outlying reads. In this study, we aim to compare inter-reader variability in LVEF measurement between 2D and 3D cardiac MRI using a single patient data set evaluated by multiple readers. We expect that 3D imaging will produce lower inter-reader variability that would lead to more precise measurement and consistent diagnoses.

Materials and

Methods: : A patient MRI dataset was utilized to measure LVEF for both 2D and 3D imaging approaches. Using a modified version of ImageJ, we segmented the LV in each of 6 long axis and 6 short axis slices. We then oriented the slices using the midseptum and using a mathematical model, generated a surface fitting mesh [Figure 1C]. This mesh enabled accurate measurement of end-systolic and end-diastolic volumes which were used to compute LVEF. For the 2D analysis, we calculated LVEF for the conventional 2 chamber and 4 chamber views. We recorded the averages and standard deviations between all 6 experienced readers for both 2D and 3D.

Results, Conclusions, and Discussions:: When measuring the standard deviation as a percentage of the mean LVEF, we found that the inter-reader variability was lowest for the 2-chamber view (5.496%), followed by the 3D view (6.033%), and lastly the 4-chamber view (7.601%). To account for 95% confidence, we took two standard deviations as our confidence intervals, ±11.0%, ±15.2%, and ±12.1% respectively.

Currently, LVEF calculated on 2D cardiac MRIs is the standard approach for diagnosing heart disease. Our confidence intervals act as thresholds for any patient’s contours analyzed by an experienced reader. If a new patient had an anticipated LVEF of 60%, but the reader discovers they have a LVEF of 52.8%, the 12% decrease in heart function could only be recognized as significant when measured under the 2 chamber long-axis view. We can say this because the 12% change in LVEF is greater than 11.0% (95% confidence interval shown in Figure 2). However, this is not the case for the 4 chamber long-axis view and the 3D view because the 12% LVEF change is not greater than either confidence interval (15.2% and 12.1% respectively).

The standard deviation calculated via the 3D MRI for this particular patient was lower than that of the 2 chamber and 4 chamber 2D MRI views, leading to a smaller variability in LVEF values. As a generalization, our readers were more precise using the 2 chamber long-axis view than the 3D view and 4 chamber long-axis view. For clinical applications, it is important to also take time commitment into consideration. Contouring for 3D cardiac MRIs takes about an hour and a half for an expert reader, while contouring for 2D cardiac MRIs takes only about ten minutes. In our future work we would like to expand the scope of our research to include more MRI readers from various experience levels so we can further test these conclusions. Additionally, we would like to add more patient datasets to better represent the overall patient population.

Acknowledgements and/or References (Optional): : Funding for this work was provided by the University of Florida Emerging Scholars Program and the Florida Breast Cancer Foundation.