Sex-specific cardiovascular risk prediction using AI-derived epicardial adipose tissue measurements on CT calcium scoring exams

Q: What problem does the study address regarding CT CAC scoring?

The study addresses that the Agatston CT-calcium scoring (CTCS) CAC score shows lower discrimination in females compared with males, motivating improved prediction strategies for women.

Q: How are epicardial adipose tissue (EAT) features used in the models?

A validated deep learning model segments EAT from CT scans, then extracts 211 fat-omics features. These features are integrated with CAC to form EAT-CAC and sex-specific EAT-CAC models.

Q: What were the main performance findings for females?

In females, CAC alone had lower discrimination than in males, while EAT-CAC improved discrimination and further improved performance using a female-specific model, with good calibration and added net benefit.

Sep 30, 2026

Uploaded by Ophelia

Sex-specific cardiovascular risk prediction using AI-derived epicardial adipose tissue measurements evaluates whether adding epicardial “fatomics” to CT coronary artery calcium (CAC) scoring improves major adverse cardiac events (MACE) prediction in females and whether sex-specific modeling provides added prognostic value. Using 40,851 individuals from the CLARIFY Registry, deep learning segmented EAT and extracted 211 fat-omics features, then Cox models compared CAC, EAT-CAC, and sex-specific EAT-CAC on a held-out test set.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Introduction

Background and clinical rationale

Methods

Study population and endpoints

Deep learning EAT segmentation and feature extraction

Model training and evaluation

Results

Discrimination and calibration

Sex-specific improvements and net benefit

Independent prediction and adjustment

Conclusions

Implications for female risk equity

Clinical Perspective

American Journal of Preventive Cardiology 25 (2026) 101367 Contents lists available at ScienceDirect American Journal of Preventive Cardiology journal [homepage:](homepage: www.journals.elsevier.com/american-journal-of-preventive-cardiology)[ www.journals.elsevier.com/american-journal-of-preventive-cardiology](homepage: www.journals.elsevier.com/american-journal-of-preventive-cardiology) | Sex-specific cardiovascular risk prediction using AI-derived epicardial adipose tissue measurements on CT calcium scoring exams | | | | | --- | --- | --- | --- | | Prerna Singha , Tao Hua, Ammar Hooria, Sadeer Al-Kindib, David L. Wilson a,c, Sanjay Rajagopalan d,*
a Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH, 44106, USA b DeBakey Heart and Vascular Center, Houston Methodist, Houston, TX 77030, USA
c Department of Radiology, Case Western Reserve University, Cleveland, OH, 44106, USA
d Harrington Heart and Vascular Institute, University Hospitals Cleveland Medical Center, Cleveland, OH, 44106, USA | | | | | A R T I C L E I N F O | | A B S T R A C T | | | Keywords:
Epicardial adipose tissue Coronary artery calcium scoring Computed tomography
Sex-specific risk prediction | | Background: The Agatston CAC score from CT-calcium scoring (CTCS) is a standard guideline recommended measure for cardiovascular risk assessment that quantifies calcified plaque burden. However, many studies have reported lower discrimination of CTCS in females. Embedded epicardial adipose tissue (EAT) features have previously been shown to improve prediction of calcium scoring, and we hypothesized that it may provide incremental prognostic information in females. In this study, we evaluate whether integrating epicardial “fatomics” with CAC score improves major adverse cardiac events (MACE) prediction from CTCS in females and assess the added value of sex-specific modeling.
Methods: 40,851 individuals undergoing CTCS for primary prevention (CLARIFY Registry, NCT04075162) were analyzed. MACE was defined as myocardial infarction, stroke, revascularization, or mortality. A validated deep learning model segmented EAT, and 211 “fat-omics” features were extracted. Cox models were trained using CAC score, CAC with fat-omics (EAT-CAC), and sex-specific EAT-CAC models trained. Performance was evaluated on a held-out test set (20 %) using C-index, calibration, and decision curves.
Results: The cohort had a mean age of 59.2 years, with 49.4 % females, and 1017 MACE events (2.49 %) over a mean follow-up of 1.7 years. On the test set (N = 8169), CAC had lower discrimination in females (C-index: 0.671) than males (0.769). EAT-CAC improved performance in females (0.691, p < 0.01 vs. CAC), with further improvement using the female-specific model (0.714, p < 0.0001 vs. CAC). No significant changes were observed in males. EAT-CAC models demonstrated good calibration, improved net benefit, and remained independently predictive after comorbidity adjustment (HR = 2.96, p < 0.001).
Conclusions: Sex-specific risk models incorporating epicardial fat-omics from CTCS improve risk prediction in females and equity in cardiovascular risk assessment. | | * Corresponding author at: University Hospitals Cleveland Medical Center, Harrington Heart & Vascular Institute, 11100 Euclid Avenue, Lakeside 5038, Cleveland, OH 44106, USA. E-mail address: [sanjay.rajagopalan@uhhospitals.org](sanjay.rajagopalan@uhhospitals.org) (S. Rajagopalan). [https://doi.org/10.1016/j.ajpc.2025.101367](https://doi.org/10.1016/j.ajpc.2025.101367) Received 25 August 2025; Received in revised form 22 October 2025; Accepted 6 December 2025 Available online 8 December 2025 2666-6677/© 2025 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC license ([http://creativecommons.org/licenses/by](http://creativecommons.org/licenses/by)nc/4.0/). P. Singh et al. American Journal of Preventive Cardiology 25 (202

FAQs

What problem does the study address regarding CT CAC scoring?

The study addresses that the Agatston CT-calcium scoring (CTCS) CAC score shows lower discrimination in females compared with males, motivating improved prediction strategies for women.

How are epicardial adipose tissue (EAT) features used in the models?

A validated deep learning model segments EAT from CT scans, then extracts 211 fat-omics features. These features are integrated with CAC to form EAT-CAC and sex-specific EAT-CAC models.

What were the main performance findings for females?

In females, CAC alone had lower discrimination than in males, while EAT-CAC improved discrimination and further improved performance using a female-specific model, with good calibration and added net benefit.

Sep 30, 2026

Uploaded by Ophelia