Machine learning in the prediction of cardiac epicardial and mediastinal fat volumes

Érick Oliveira Rodrigues, V. H. A. Pinheiro, Panos Liatsis, Aura Conci

Introduction

Cardiac fat deposits, specifically epicardial fat (located inside the pericardial sac) and mediastinal fat (located outside the sac), are recognized as significant biomarkers for cardiovascular and metabolic risks, including coronary artery calcification. Accurate quantification of these volumes from Computed Tomography (CT) scans is crucial for patient risk assessment. However, the current standard of manual or semi-automated segmentation presents a major clinical bottleneck; manual segmentation is tedious, and even automated methods can take nearly two hours per patient.

Anatomical Context and the Prediction Hypothesis

Understanding the spatial relationship between these fat deposits is key to the study's premise. Epicardial fat is separated from mediastinal fat by the pericardium. Because these two fat deposits are anatomically adjacent and highly correlated, researchers investigated whether it is possible to "skip" the segmentation of one by predicting its volume based on the other, thereby streamlining the diagnostic pipeline.

Anatomical layers of the heart, illustrating the relationship between the pericardium and the epicardium (visceral layer of serous pericardium).

Methodology and Data

The study utilized a ground-truth dataset consisting of CT scans from 20 patients, totaling 878 axial-plane images. Experts manually segmented these images to identify epicardial fat, mediastinal fat, and the pericardium. This high-quality annotation allowed the researchers to frame volume estimation as a regression problem.

Expert segmentation of CT slices showing epicardial fat (red), mediastinal fat (green), and the pericardium (blue).

Seven features were extracted from each CT slice, including pixel counts for tissue types and slice indices. The authors evaluated a wide array of machine learning algorithms using the Weka framework, including meta-algorithms like Rotation Forest, Neural Networks (MLP), and interpretable Linear Regression models.

Key Results and Efficiency Gains

The "Segment One, Predict One" strategy proved highly effective. The top-performing model, a Rotation Forest with an MLP regressor, achieved a Pearson correlation coefficient (ρ) of 0.9876 when predicting mediastinal fat from epicardial fat, with a Relative Absolute Error (RAE) of only 14.4%.

Key insights from the results include:

  • Efficiency: By segmenting only one fat deposit and predicting the second, total processing time was reduced from 1.8 hours to approximately 0.9 hours—a 50% speedup.
  • Asymmetric Predictability: It is slightly easier to predict mediastinal fat from epicardial fat (ρ = 0.9876) than vice versa (ρ = 0.9683).
  • Clinical Utility: Even simple Linear Regression performed remarkably well (ρ ≈ 0.95), providing clinicians with straightforward functional models for manual estimation.

Conclusion and Future Directions

By leveraging the high correlation between adjacent cardiac fat deposits, machine learning models can significantly streamline the diagnostic pipeline without significant loss in accuracy. While the study relied on a relatively small cohort of 20 patients, the results demonstrate a clear path toward faster cardiac risk assessments. Future work will focus on clinical validation with larger, more diverse datasets and testing generalization across different CT scanner manufacturers.

Summary updated

Abstract

We propose a methodology to predict the cardiac epicardial and mediastinal fat volumes in computed tomography images using regression algorithms. The obtained results indicate that it is feasible to predict these fats with a high degree of correlation, thus alleviating the requirement for manual or automatic segmentation of both fat volumes. Instead, segmenting just one of them suffices, while the volume of the other may be predicted fairly precisely. The correlation coefficient obtained by the Rotation Forest algorithm using MLP Regressor for predicting the mediastinal fat based on the epicardial fat was 0.9876, with a relative absolute error of 14.4% and a root relative squared error of 15.7%. The best correlation coefficient obtained in the prediction of the epicardial fat based on the mediastinal was 0.9683 with a relative absolute error of 19.6% and a relative squared error of 24.9%. Moreover, we analysed the feasibility of using linear regressors, which provide an intuitive interpretation of the underlying approximations. In this case, the obtained correlation coefficient was 0.9534 for predicting the mediastinal fat based on the epicardial, with a relative absolute error of 31.6% and a root relative squared error of 30.1%. On the prediction of the epicardial fat based on the mediastinal fat, the correlation coefficient was 0.8531, with a relative absolute error of 50.43% and a root relative squared error of 52.06%. In summary, it is possible to speed up general medical analyses and some segmentation and quantification methods that are currently employed in the state-of-the-art by using this prediction approach, which consequently reduces costs and therefore enables preventive treatments that may lead to a reduction of health problems.
BibTeX
@misc{rodrigues2022machine,
  author = {Érick Oliveira Rodrigues and V. H. A. Pinheiro and Panos Liatsis and Aura Conci},
  title = {{Machine learning in the prediction of cardiac epicardial and mediastinal fat volumes}},
  year = {2022},
  eprint = {2208.14374},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2208.14374},
}
Figure