Cardiac visceral fat volume estimation with machine learning: an algorithm for screening ultra-low-dose computed tomography

Congress:

ECR 2020

Poster Number:

C-10054

Type:

Scientific Exhibit

Keywords:

Case-control study, Retrospective, Image registration, Arteriosclerosis, Segmentation, Screening, CAD, CT, Computer applications, Cardiac, Artificial Intelligence, Artificial Intelligence and Machine Learning, Performed at one institution

Authors:

V. Chernina¹, M. Pisov¹, M. Belyaev¹, V. Gombolevskiy¹, I. Blokhin¹, I. Bekk², O. Aleshina¹, T. Korb¹, S. Morozov¹; ¹Moscow/RU, ²Moscow /RU

DOI:

10.26044/ecr2020/C-10054

DOI-Link:

https://dx.doi.org/10.26044/ecr2020/C-10054

Fig. 2: Materials and methods. AI training dataset inclusion and exclusion criteria.

Fig. 3: Materials and methods. AI validation dataset inclusion and exclusion criteria.

Fig. 4: Graphic user interface for pericardial and mediastinal contour tagging and...

Fig. 5: Time required for the semiautomatic tagging of epicardial adipose tissue .

Purpose

White adipose tissue can be divided into visceral and subcutaneous adipose tissue. The former surrounds various organs and spaces, including the omentum, mesentery, mediastinum, and perivascular regions [1]. Two functionally different adipose tissue types - epicardial adipose tissue (EAT) and paracardial adipose tissue (PAT) - surround almost all arteries and the heart ( Fig. 1 ) [2]. According to the Multi-Ethnic Study of Atherosclerosis (MESA), the volume of pericardial adipose tissue is an independent predictor of ischemic heart disease (IHD) [3]. The most exciting is...

Methods and materials

The sample size was 500 patients (men-to-women ratio, 1:1) ( Fig. 2 ). Random recruitment was carried out in 5 age groups (from 50 to 75 years), a hundred patients in each age group, equally smokers and non-smokers. Additionally, we randomly selected 100 cases (equally ULDCT and CT) to test robustness ( Fig. 3 ). Two radiologists with 5-yearexperience in thoracic radiology manually tagged each study. The fat density threshold was from -190 HU to -30 HU. The robustness of measurements was assessed with Syngo.via....

Results

We created software for data tagging and export ( Fig. 4 ). Average tagging time for 1 study was 30 minutes ( Fig. 5 ) A deep learning-based method estimated pericardial contour. First, a shallow network localizes the heart. Next, the region is converted to cylindrical coordinates, and another network determines pericardial contour. We measured the inter-expert variability via average symmetric surface distance (ASSD). The mean ASSD was 1.88±0.65. The ASSD between the experts and the algorithm was 1.81±0.54 ( Fig. 6 ). Mean radiologist...

Conclusion

The deep learning algorithm for epicardial fat volumetry in ULDCT showed promising results. ML-based volume estimation is feasible and may be used as a predictor of preclinical coronary heart disease.

Personal information and conflict of interest

V. Chernina; Moscow/RU - nothing to disclose M. Pisov; Moscow/RU - nothing to disclose M. Belyaev; Moscow/RU - nothing to disclose V. Gombolevskiy; Moscow/RU - nothing to disclose I. Blokhin; Moscow/RU - nothing to disclose I. Bekk; Moscow/RU - nothing to disclose O. Aleshina; Moscow/RU - nothing to disclose T. Korb; Moscow/RU - nothing to disclose S. Morozov; Moscow/RU - CEO at Research and practical clinical center for diagnostics and telemedicine technologies of Moscow Health Care Department

References

1. Bjorndal B., Burri L., Staalesen V., Skorve J., Berge R.K. Different adipose depots: their role in the development of metabolic syndrome and mitochondrial response to hypolipidemic agents. J. Obes. 2011; 2011: 1-15. DOI: 10.1155/2011/490650 2. Iacobellis G., Corradi D., Sharma A.M. Epicardial adipose tissue: anatomic, biomolecular and clinical relationships with the heart. Nat. Clin. Pract. Cardiovasc. Med. 2005; 2: 536-43. DOI: 10.1038/ ncpcardio0319 3. Ding J., Hsu F.C., Harris T.B., Liu Y., Kritchevsky S.B., Szklo M. et al. The association of pericardial fat with...