Evaluation of a AI-based plane prescription tool in cardiac MRI (RF_SA_497)

Saturday, February 1, 2025

11:30 AM – 11:40 AM East Coast USA Time

Location: Blue Room PreFunction

Presenting Author(s)

BB

Benjamin Böttcher, MD

Resident / Physician
Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology / University Medical Center Rostock, Germany

Primary Author(s)

AK

Ann-Christin Klemenz, PhD

Postdoctoral Researcher
Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology / University Medical Center Rostock, Germany

Co-Author(s)

KD

Karolin Deyerberg, MD

Doctoral candidate / Student
Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology / University Medical Center Rostock, Germany
LW

Lena-Maria Watzke, MD

Doctoral candidate / Student
Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology / University Medical Center Rostock, Germany
MG

Margarita Gorodezky, PhD

Clinical Scientist
GE HealthCare, Germany
MM

Mathias Manzke

Scientist
Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology / University Medical Center Rostock, Germany
MW

Marc-André Weber, MD, MSc

Institute Director
Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology / University Medical Center Rostock, Germany
FM

Felix G. Meinel, MD

Vize Chair
Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology / University Medical Center Rostock, Germany

Background: Cardiac MRI provides a comprehensive evaluation of the heart and is considered the gold standard for the assessment of cardiac function hence playing a pivotal role in diagnostics of various cardiac diseases. Volumetric analysis of the right and left ventricle is performed on cine sequences in well-established standard imaging planes such as the short axis stack, 2-, 3-, and 4-chamber views to ensure reliability and comparability of the results. Correct visualization and angulation of these planes is therefore absolutely necessary. Manual planning of the standard cardiac imaging planes is time-consuming and requires special training of technicians and radiologists. Moreover, image plane quality is impacted by the experience of the planning individual. Artificial intelligence (AI) based applications could accelerate und simplify this step and make cardiac MRI available for facilities with lower availability of highly specialized staff.

Methods: In a prospective cohort study with 25 healthy volunteers (13 female, median 40 (23 to 64) years old) a fully automated plane positioning tool for cardiac MRI was compared to expert manual planning on a 1.5T MRI scanner (Signa Artist, GE Healthcare). Following cardiac cine imaging planes were assessed: a short-axis stack (Sax), 2-, 3- and 4-chamber view (FOV: 34x34cm, Matrix size: 200x224, slice thickness: 8mm, slice spacing: 2mm). To compare the performance between manual and fully automated plane prescription, left ventricular function (end diastolic volume – EDV; end systolic volume – ESV; stroke volume – SV; ejection fraction - EF) of both planning methods was evaluated from the short axis stack using cvi42 (Circle Cardiovascular Imaging, Calgary). Additionally, a biplane volumetric analysis from 2- and 4-chamber cine image series was performed in order to evaluate the potential impact of angle discrepancies in the long axis imaging planes. A Wilcoxon matched-pairs signed rank test and Bland-Altman plots were used for statistical analysis.

Results: EDV, ESV, SV and EF from biplane and ESV, SV and EF from Sax volumetric analysis showed no significant difference between manual and fully automated plane positioning. EDV calculated from the Sax stack showed a statistically significant difference between manual and AI based slices prescription (p=0.0184). However, the absolute difference was very small and not clinically significant (mean bias -1.7mL). Details of the Bland-Altman analysis can be found in Table 1. Bland-Altman graphs of short axis volumetry of manual and AI based plane positioning is displayed in Figure 1. An image example is shown in Figure 2.

Conclusion: AI assisted cardiac slice prescription provides high quality and accurate images without clinically relevant impact on volumetric assessment. AI based cardiac planning has the potential to provide more robust data than manual slice prescription. In future this could significantly decrease overall cardiac MRI exam time and make cardiac MRI available for a broader range of patients.