Epicardial fat Tissue Volumetry: Comparison of Semi-Automatic Measurement and the Machine Learning Algorithm

Valeria Chernina, Maxim Pisov, Mikhail Belyaev, I. V. Bekk, K. A. Zamyatina, Т. А. Корб, Olga Aleshina, E. A. Shukina, A. V. Solovev, R. A. Skvortsov, D. A. Filatova, Dmitry Sitdikov, Anna Chesnokova, С. П. Морозов, V. A. GombolevskyKardiologiia 2020 · DOI

Abstract

Aim: To compare assessments of epicardial adipose tissue (EAT) volumes obtained with a semi-automatic, physician-performed analysis and an automatic analysis using a machine-learning algorithm by data of low-dose (LDCT) and standard computed tomography (CT) of chest organs.

Material and methods: This analytical, retrospective, transversal study randomly included 100 patients from a database of a united radiological informational service (URIS). The patients underwent LDCT as a part of the project "Low-dose chest computed tomography as a screening method for detection of lung cancer and other diseases of chest organs" (n=50) and chest CT according to a standard protocol (n=50) in outpatient clinics of Moscow. Each image was read by two radiologists on a Syngo.via VB20 workstation. In addition, each image was evaluated with a developed machine-learning algorithm, which provides a completely automatic measurement of EAT.

Results: Comparison of EAT volumes obtained with chest LDCT and CT showed highly consistent results both for the expert-performed semi-automatic analyses (correlation coefficient >98 %) and between the expert layout and the machine-learning algorithm (correlation coefficient >95 %). Time of performing segmentation and volumetry on one image with the machine-learning algorithm was not longer than 40 sec, which was 30 times faster than the quantitative analysis performed by an expert and potentially facilitated quantification of the EAT volume in the clinical conditions.

Conclusion: The proposed method of automatic volumetry will expedite the analysis of EAT for predicting the risk of ischemic heart disease.

Introduction

Epicardial adipose tissue (EAT)—the fat surrounding the heart—is a potent independent predictor of coronary heart disease. While its volume can be measured via CT scans, doing so is a significant bottleneck in clinical workflows. Currently, radiologists must use semi-automatic tools that require manual tracing of the pericardium, taking roughly 15 minutes per patient. This labor-intensive process makes EAT quantification impractical for large-scale screening. The authors sought to bridge this gap by creating a fully automated AI solution that works reliably on both standard CT scans and low-dose CT (LDCT) scans typically used in lung cancer screening programs.

Methodology

The researchers utilized data from the Moscow Unified Radiological Information Service (URIS). The machine learning (ML) algorithm was trained on a substantial dataset of 352 LDCT scans and 97 standard CT scans, then validated on a separate set of 88 LDCT and 25 standard CT scans. For the final head-to-head comparison, 100 patients were randomly selected (50 standard CT, 50 LDCT).

Patient exclusion criteria for the study cohort.

The study compared two primary measurement approaches:

  1. Semi-automatic: Two radiologists performed measurements on a Syngo.via VB20 workstation (Siemens Healthineers).
  2. Automatic: A novel ML pipeline using 3D Convolutional Neural Networks (CNNs) localized the heart and segmented the pericardial contour. To ensure a continuous and anatomically correct boundary, the system converts the image into cylindrical coordinates, predicting the pericardial contour in this space before mapping it back to standard coordinates.
Comparative workflow between semi-automatic radiologist assessment and the machine learning algorithm.

Key Results

  • High Accuracy: The correlation between the AI and human experts was >95%. For context, the correlation between two different human experts was >98%, meaning the AI is performing very close to the "gold standard."
  • Massive Speedup: The AI completed the task in 38 ± 2 seconds, compared to 14–17 minutes for a radiologist. This represents a 30x increase in efficiency.
  • Consistency: The difference in volume measurements between the AI and humans averaged less than 9 ml (roughly 5-6% of total volume), which is clinically negligible.

Discussion

One of the most significant advantages of this model is the use of cylindrical coordinates. Standard segmentation often struggles with "leaky" boundaries where the pericardium is thin. By predicting the boundary as a curve in cylindrical space, the algorithm guarantees a closed, continuous loop around the heart. Furthermore, the AI remained highly accurate even on noisier LDCT scans, suggesting it can effectively handle the graininess that often complicates manual human tracing.

While the results are promising, future work is needed to validate the model on larger, more diverse cohorts and to refine the algorithm for "ultra-low dose" CT environments where noise is even more prevalent.

Conclusion

The proposed method of automatic volumetry significantly expedites the analysis of EAT. By reducing processing time from 15 minutes to under 40 seconds, EAT measurement can move from a research curiosity to a standard component of radiology reports, facilitating better prediction of ischemic heart disease risk in clinical practice.

Summary updated

Abstract

Aim To compare assessments of epicardial adipose tissue (EAT) volumes obtained with a semi-automatic, physician-performed analysis and an automatic analysis using a machine-learning algorithm by data of low-dose (LDCT) and standard computed tomography (CT) of chest organs.Material and methods This analytical, retrospective, transversal study randomly included 100 patients from a database of a united radiological informational service (URIS). The patients underwent LDCT as a part of the project "Low-dose chest computed tomography as a screening method for detection of lung cancer and other diseases of chest organs" (n=50) and chest CT according to a standard protocol (n=50) in outpatient clinics of Moscow. Each image was read by two radiologists on a Syngo. via VB20 workstation. In addition, each image was evaluated with a developed machine-learning algorithm, which provides a completely automatic measurement of EAT.Results Comparison of EAT volumes obtained with chest LDCT and CT showed highly consistent results both for the expert-performed semi-automatic analyses (correlation coefficient >98 %) and between the expert layout and the machine-learning algorithm (correlation coefficient >95 %). Time of performing segmentation and volumetry on one image with the machine-learning algorithm was not longer than 40 sec, which was 30 times faster than the quantitative analysis performed by an expert and potentially facilitated quantification of the EAT volume in the clinical conditions.Conclusion The proposed method of automatic volumetry will expedite the analysis of EAT for predicting the risk of ischemic heart disease.
BibTeX
@inproceedings{chernina2020epicardial,
  author = {Valeria Chernina and Maxim Pisov and Mikhail Belyaev and I. V. Bekk and K. A. Zamyatina and Т. А. Корб and Olga Aleshina and E. A. Shukina and A. V. Solovev and R. A. Skvortsov and D. A. Filatova and Dmitry Sitdikov and Anna Chesnokova and С. П. Морозов and V. A. Gombolevsky},
  title = {{Epicardial fat Tissue Volumetry: Comparison of Semi-Automatic Measurement and the Machine Learning Algorithm}},
  booktitle = {Kardiologiia},
  year = {2020},
  doi = {10.18087/cardio.2020.9.n1111},
}
Figure