Hierarchical Consensus Clustering of CMR Radiomics Reveals Important Histopathological Differences in Nonischemic Dilated Cardiomyopathy (RF_FR_351)

Friday, January 31, 2025

11:20 AM – 11:30 AM East Coast USA Time

Location: Blue Room PreFunction

Presenting Author(s)

Amine Amyar, PhD

Instructor in medicine
Harvard Medical School - Beth Israel Deaconess Medical Center

Primary Author(s)

Amine Amyar, PhD

Instructor in medicine
Harvard Medical School - Beth Israel Deaconess Medical Center

Co-Author(s)

SN

Shiro Nakamori, MD, PhD

Visiting Scientist
Beth Israel Deaconess Medical Center
LN

Long H. Ngo, PhD

Associate Professor
Harvard Medical School
Masaki Ishida, MD, PhD

Associate Professor
Mie University Hospital, Japan
SN

Satoshi Nakamura, MD, PhD

Associate Professor
Mie University Hospital, Japan
TO

Taku Omori, MD, PhD

Doctor
Mie University Hospital, Japan
KM

Keishi Moriwaki, MD

Doctor
Mie University Hospital, Japan
NF

Naoki Fujimoto, MD

Doctor
Mie University Hospital, Japan
KI

Kyoko Imanaka-Yoshida, MD

Doctor
Mie University Hospital, Japan
Hajime Sakuma, MD

Professor and Chariman
Mie University, Japan
KD

Kaoru Dohi, MD, PhD

Professor
Mie University Hospital, Japan
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

Background: A hallmark of non-ischemic dilated cardiomyopathy (DCM) is the heterogeneity of its histologic subtypes that results in different responses to medical therapy and diverse patient outcomes. While endomyocardial biopsy (EMB) provides more accurate phenotyping, it is an invasive procedure that carries risks. In this study, we explore the potential of CMR radiomics hierarchical consensus clustering to non-invasive phenotyping in DCM patients with distinct histological characteristics.

Methods: We retrospectively identified 132 DCM patients (54±15 years; 71% male) who underwent 3T CMR (Ingenia, Philips, Netherlands) within 6 [2-5] days prior to their scheduled EMB. The CMR imaging protocol included native and post-contrast T₁ for extracellular volume (ECV) mapping, and late gadolinium enhancement (LGE). Radiomic first-order and textural features were computed for each sequence from the mid-septal myocardium near the biopsy region (1023 features). Hierarchical consensus clustering, an unsupervised data-driven approach, was then applied to identify distinct patient clusters by identifying the optimal number of clusters without relying on specific ground truths or outcomes. These clusters were then compared based on the histological features of the septal biopsy samples. An unpaired t-test was used for comparisons between two clusters, while a one-way ANOVA with Bonferroni adjustment for multiple comparisons was used when there were more than two clusters.

Results:

Four clusters of patients were identified using native T₁ radiomic features. Cluster 1 was characterized by a lower collagen volume fraction (CVF) of 10.9% and a lower prevalence of LGE (43%) with inflammation present in 14% of patients. Cluster 2 showed a higher prevalence of inflammation (27%, p=0.03), increased extracellular space (30.6%, p=0.009), and a greater presence of LGE (66%, p=< 0.001). Cluster 3 had a lower prevalence of LGE (41%) without ongoing inflammation, while Cluster 4 had a higher prevalence of replacement fibrosis (30%, p=0.014) and trend for a higher left ventricular mass (p=0.072). Notably, there were no significant differences in native T₁ among the clusters (p=0.545). ECV radiomics identified two clusters with distinct underlying histological profiles: Cluster 1 was characterized by a severe myocardial fibrosis (20%, p=0.013), myocyte hypertrophy (cell diameter standard deviation: 3.7±1.7, p=0.006), and inflammation (16%, p=0.067), while Cluster 2 demonstrated less fibrosis (CVF of 10.5%, p=0.045) and reduced inflammation (2%). LGE radiomics also showed potential for phenotyping: Cluster 1 showed inflammation (25%, p=0.081) along with vacuolization (25%, p=0.098); Cluster 2 consisted of patients with minimal CVF (11%, p=0.137); Cluster 3 had low CVF (12%) but immune cell infiltrates (10%); and Cluster 4 demonstrated severe myocardial fibrosis (42%, p=0.001).

Conclusion: Hierarchical consensus clustering of radiomic features extracted from mid septal native T₁, ECV, and LGE revealed important histopathological differences in DCM, potentially enabling more precise non-invasive phenotyping.