Leveraging calcium score CT radiomics for heart failure risk prediction

Area of Science:

• Cardiology
• Radiology
• Artificial Intelligence

Background:

• Predicting heart failure (HF) risk traditionally relies on extensive clinical data.
• Diabetes mellitus (DM) is a significant risk factor for HF, necessitating tailored risk assessment strategies.

Purpose of the Study:

• To develop and validate a screening method for HF risk prediction using only computed tomography calcium scoring (CTCS).
• To assess the utility of radiomic features from calcifications ("calcium-omics") and epicardial adipose tissue ("fat-omics") for HF risk stratification.

Main Methods:

• Analysis of CTCS scans from 1,998 patients (including 336 with type 2 diabetes) from the CLARIFY Study registry.
• Deep learning for epicardial adipose tissue (EAT) segmentation and radiomic feature engineering for "calcium-omics" and "fat-omics".
• Development and comparison of predictive models incorporating radiomics, clinical factors, EAT volume, and Agatston score.

Main Results:

• In the overall cohort, "fat-omics" models demonstrated superior HF prediction (C-index: 77.3) compared to clinical factors and other radiomic approaches.
• For patients with DM, the "calcium-omics" model achieved the highest predictive performance (C-index: 81.8).
• CTCS-based radiomic models significantly outperformed traditional clinical prediction scores.

Conclusions:

• CTCS-based radiomic models, particularly "fat-omics" and "calcium-omics", offer a powerful tool for predicting incident HF.
• These CTCS-derived models provide enhanced HF risk stratification compared to conventional clinical assessments.
• The findings support the use of CTCS for non-invasive HF screening, with tailored approaches for diabetic and non-diabetic populations.