CMR-Analysis (including machine learning)

1354428 - Myocardial Strain is Reduced in Hypertrophic Segments in Patients with Hypertrophic Cardiomyopathy – Retrospective MRI Analysis with Deep Learning Synthetic Strain

Wednesday, January 25, 2023

Location: E-POSTER

Presenting Author(s)

AS

Albert Song, PhD

Medical Student
University of California, San Diego, United States

Primary Author(s)

AS

Albert Song, PhD

Medical Student
University of California, San Diego, United States

Co-Author(s)

EM

Evan Masutani, PhD

Medical Student
University of California, San Diego, United States
RC

Rahul Chandrupatla

Medical Student
University of California, San Diego, United States
EW

Elizabeth Weihe, MD

Associate Clinical Professor
University of California, San Diego, United States
LH

Lewis Hahn, MD

Assistant Professor
University of California, San Diego, United States
JS

Jorge Silva Enciso, MD

Associate Professor, Director of HCM Program
University of California, San Diego, United States
Albert Hsiao, MD, PhD

Associate Professor
UC San Diego
La Jolla, California, United States

Background: Hypertrophic cardiomyopathy (HCM) is an autosomal dominant disorder of cardiac sarcomeric proteins, characterized by left ventricular (LV) hypertrophy, affecting 1 in 500 adults. These sarcomeric defects cause myocyte disarray and interstitial fibrosis which compromise myocardial function and can progress to heart failure. While late gadolinium enhancement is the current gold standard for non-invasive assessment and prognostication, it has been suggested that global myocardial strain may provide additional prognostic value in HCM (1). We recently developed a deep learning synthetic strain (DLSS) algorithm, which has potential to retrospectively measure myocardial strain with regional granularity from routine cine SSFP images.

Purpose: To evaluate whether a novel MRI strain algorithm, DLSS, can quantitatively identify regional impairment of myocardial shortening in hypertrophic segments in HCM patients.

Methods: DLSS was trained to infer myocardial velocity fields for calculation of strain from cine SSFP images, using myocardial velocities from 4D Flow MRI obtained from 223 cardiac MRIs. We retrospectively curated a separate, independent cohort of 55 patients with HCM, for whom DLSS was used to calculate segmental radial and circumferential myocardial strain and compared with end-diastolic wall thickness. Myocardial segments were partitioned according to wall thickness: normal (< 12 mm), moderately hypertrophic (12-15 mm), and severely hypertrophic ( >15 mm) (2,3). Statistical analyses included one-way analysis of variance (ANOVA) and Student’s t-test, both with a type I error threshold of 0.05 (p< 0.05).

Results: Across 55 HCM patients, end-diastolic wall thickness was inversely correlated with radial strain (r=-0.36, p< 0.001) and directly correlated with circumferential strain (r=0.30, p< 0.001). There was significantly reduced radial and circumferential strain in hypertrophic segments (p< 0.001). Median radial strain of normal, moderately hypertrophic, and severely hypertrophic segments were 43.9% (IQR 34 to 54.7%), 32.6% (IQR 23.6 to 41.9%), and 25.3% (IQR 12.2 to 35.9%), respectively. Median circumferential strain of normal, moderately hypertrophic, and severely hypertrophic ( >12 mm) segments were -42.9% (IQR -54.1 to -33.2%), -32.7% (IQR -42.9 to -25%) and -19.4% (IQR -31.9 to -13.8%), respectively.

Conclusion: Retrospective algorithms for measurement of myocardial strain have potential to reveal insights from historical MRI data. In patients with HCM, we observe regional reductions in myocardial strain among hypertrophic segments. New deep learning algorithms for analysis of strain may complement other non-contrast MRI methods for myocardial tissue characterization, and have potential to improve risk stratification.