Free-Breathing, High Spatiotemporal Resolution, Single-Beat Exercise-CMR Cine with Generative AI Image Enhancement (RF_TH_160)

Presenting Author(s)

• 

  Fahime Ghanbari, MD

  Fellow
  Harvard medical school, Beth Israel Deaconess Medical Center

Primary Author(s)

• 

  Fahime Ghanbari, MD

  Fellow
  Harvard medical school, Beth Israel Deaconess Medical Center

Co-Author(s)

• 

  Alexander Schulz, MD

  Dr.
  Harvard Medical School / BIDMC, Germany
• 

  Manuel Morales, PhD

  Instructor of Medicine
  Beth Israel Deaconess Medical Center and Harvard Medical School
• JR

  Jennifer Rodriguez

  Clinical Trials Specialist
  Beth Israel Deaconess Medical Center
• CH

  Christopher Hoeger, MD

  Instructor
  Beth Israel Deaconess Medical Center
• CT

  Connie Tsao, MD, MPH

  Physician
  Beth Israel Deaconess Medical Center
• WM

  Warren J. Manning, MD

  Section Chief, Non-invasive Cardiac Imaging
  Beth Israel Deaconess Medical Center/Harvard Medical School
  Beth Israel Deaconess
• RN

  Reza Nezafat, PhD

  Professor
  Harvard Medical School

Figure 1 (A): Free-breathing, single-beat cine acquisition collects data for each slice in one heartbeat. A dummy heartbeat is used to take the magnetization to steady-state. (B): Image reconstruction pipeline with non-uniform k-space sampling. Low-resolution images are enhanced with the REGAIN. (C): Exercise protocol.

Figure 2. Example of rest and stress imaging using single-beat cine acquisition in the same subject, illustrating three representative slices acquired from the basal, mid-ventricular, and apical regions.


Background: Exercise CMR (Ex-CMR) can reveal cardiac dysfunction that is not evident at rest. However, Ex-CMR is technically challenging, requiring high spatiotemporal resolution, non-ECG-segmented acquisition, and free breathing. Accurate volume and function calculations need full heart coverage in under one minute. Thus, we sought to develop a free-breathing, ECG-triggered, single-beat cardiac cine with generative AI image enhancement to quantify cardiac functional reserve with Ex-CMR.

Methods: A single-beat sequence was implemented, collecting data for each slice in one heartbeat (Fig.1A). The sequence collects data for each slice within a heartbeat before moving to the adjacent slice. It acquires a dummy heartbeat before each slice to achieve steady-state magnetization. Fig. 1B shows the pipeline for image acquisition/reconstruction. Images are prescribed at a lower spatial resolution along the phase-encode direction with a compressed sensing (CS) acceleration rate of 7.4. A vendor-provided CS reconstruction is used to reconstruct low-resolution images. The low-resolution images are then input to resolution enhancement generative adversarial inline network (REGAIN) to improve image sharpness. Images were collected with a spatiotemporal resolution of 1.9×1.9 mm² and 37ms. Ex-CMR was performed on 40 healthy adults using a supine ergometer, with imaging at rest and immediately post-exercise (Fig.1C). 15 were re-imaged one year later to assess scan-rescan reproducibility. Rest and stress biventricular parameters (ESV, EDV, SV, EF) and their corresponding Δ (%) (cardiac functional reserve) were calculated using CVI42. 50% of the data were randomly selected for interobserver analysis, with repeat scans independently analyzed by two readers. 4 readers assessed diagnostic and image quality using a three-point scale: diagnostic (1), non-confident (2), and non-diagnostic (3) for diagnostic quality; good (1), adequate (2), and poor (3) for image quality. Adequate quality mainly impacted epicardial contouring, while poor quality affected both contours. Statistical analysis included ICC.

Results: Fig. 2 shows example images. LVEDV increased by 11±3%, LVESV decreased by 24±6%, leading to a 37±6% increase in LVSV and a 23±5% improvement in LVEF. RVEDV increased by 7±4%, RVESV decreased by 19±9%, leading to a 30±9% rise in RVSV and a 22±8% improvement in RVEF. Interobserver analysis showed good to excellent agreement. Scan-rescan reproducibility was also good to excellent across readers (Table 1). Readers 1, 2, and 3 rated stress images as diagnostic in 98%, 98%, and 83% of cases, respectively, while Reader 4 rated 45% as diagnostic and 50% as non-confident. Endocardial border tracing was deemed feasible (good/adequate) in 100%, 100%, 98%, and 98% of images by Readers 1, 2, 3, and 4, respectively.

Conclusion: A 14.8-fold accelerated, high spatiotemporal resolution single-beat Ex-CMR cine with generative AI image enhancement enables reproducible functional reserve quantification.

Figure 1 (A): Free-breathing, single-beat cine acquisition collects data for each slice in one heartbeat. A dummy heartbeat is used to take the magnetization to steady-state. (B): Image reconstruction pipeline with non-uniform k-space sampling. Low-resolution images are enhanced with the REGAIN. (C): Exercise protocol.

Figure 2. Example of rest and stress imaging using single-beat cine acquisition in the same subject, illustrating three representative slices acquired from the basal, mid-ventricular, and apical regions.