Radiomics of Late Gadolinium Enhancement in Pediatric Hypertrophic Cardiomyopathy with Ventricular Arrhythmia

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)

• 

  Lars Grosse-Wortmann, MD

  Professor of Pediatrics
  Oregon Health & Science University
• RC

  Raymond H. Chan, MD, MPH

  Cardiologist
  University Health Network, Canada
• LN

  Long H. Ngo, PhD

  Associate Professor
  Harvard Medical School
• SB

  Seshadri Balaji, MD, PhD

  pediatric cardiologist
  Oregon Health and Science University
• RN

  Reza Nezafat, PhD

  Professor
  Harvard Medical School

Background: Sudden cardiac death (SCD) is a devastating event in young patients with hypertrophic cardiomyopathy (HCM). The presence and extent of late gadolinium enhancement (LGE) have been shown to have prognostic value in adult (1) and pediatric HCM (2). While scar heterogeneity assessed by radiomics on LGE has shown prognostic value in adult HCM (3), its utility in identifying high-risk patients in pediatric HCM remains unknown. This study aims to determine whether LGE radiomics can identify high-risk SCD patients in pediatric HCM.

Methods: In a cross-sectional retrospective study, we identified pediatric HCM patients with SCD events from a multi-center database of pediatric HCM patients across 37 international sites. SCD events were defined as a composite of SCD, appropriate implantable cardioverter-defibrillator shock, or aborted cardiac arrest. All patients underwent CMR with LGE and had available follow-up data. We randomly selected age and sex-matched cases from the remaining cohort without SCD events during follow-up. CMR imaging was performed using either a 1.5T or 3T scanner.  First-order and textural radiomic features, reflecting increasing LGE complexity and heterogeneity, were extracted from the left ventricular myocardium using short-axis LGE images from each slice, including both healthy and scarred myocardium when present. This approach eliminated the need for LGE scar quantification and provided a comprehensive representation of the myocardium within the LGE images. We calculated the radiomics mean value across short-axis slices for each patient. Highly correlated features (correlation coefficient >0.9) were removed to avoid redundancy. Principal component analysis was then applied for unsupervised dimensionality reduction, yielding three principal radiomic components (referred to as PrinRads). The stratification power of the three LGE PrinRads was assessed in a logistic regression model (Figure 1). The model's performance was evaluated using the area under the receiver-operating curve (AUC).

Results: Table 1 summarizes the clinical and demographic characteristics of the study cohort, along with the CMR findings in the SCD and non-SCD groups. Sixty-eight HCM pediatric patients were included in the study (72% male, 14.4±4.9 years), including 34 with SCD. LGE PrinRads successfully stratified patients into high- and low-risk groups with an AUC of 0.76 (p< 0.001) (Figure 2A). Moreover, one principal radiomic (PrinRad1) was independently associated with SCD (Odds ratio: 0.49 [95% confidence interval (CI): 0.25 to 0.90]), even after correcting for the presence and extent of LGE. The mean value of PrinRad1 was significantly lower in the SCD group compared to the group without events (-0.41 [95% CI: -0.71 to -0.10] vs. 0.41 [95% CI: 0.07 to 0.74]; p< 0.001) (Figure 2B).

Conclusion: In this first proof-of-concept study on radiomics in pediatric HCM patients, LGE radiomics was associated with an increased risk of HCM-related SCD. Further research will be geared towards better discrimination of patients with and without SCD.

Figure 1. The analytical framework employed for extracting late gadolinium enhancement (LGE) radiomics to detect patients at risk (A). The sequence for computing radiomic features from input LGE short-axis slices is outlined in (B).

Figure 2. Receiver-Operating Characteristics curve for the risk stratification model of sudden cardiac death (A), and scatter plot with mean ± 95% confidence interval (CI) for principal component radiomic 1 (PrinRad1) (B). The mean and distribution of the PrinRad1 are different between patients with and without sudden cardiac death.