Artificial intelligence based automatic quantification of epicardial adipose tissue suitable for large scale population studies

Abstract

To develop a fully automatic model capable of reliably quantifying epicardial adipose tissue (EAT) volumes and attenuation in large scale population studies to investigate their relation to markers of cardiometabolic risk. Non-contrast cardiac CT images from the SCAPIS study were used to train and test a convolutional neural network based model to quantify EAT by: segmenting the pericardium, suppressing noise-induced artifacts in the heart chambers, and, if image sets were incomplete, imputing missing EAT volumes. The model achieved a mean Dice coefficient of 0.90 when tested against expert manual segmentations on 25 image sets. Tested on 1400 image sets, the model successfully segmented 99.4% of the cases. Automatic imputation of missing EAT volumes had an error of less than 3.1% with up to 20% of the slices in image sets missing. The most important predictors of EAT volumes were weight and waist, while EAT attenuation was predicted mainly by EAT volume. A model with excellent performance, capable of fully automatic handling of the most common challenges in large scale EAT quantification has been developed. In studies of the importance of EAT in disease development, the strong co-variation with anthropometric measures needs to be carefully considered.

Pre-diabetes is Associated with Attenuation rather than Volume of Epicardial Adipose Tissue on Computed Tomography

“…It is clear that both BMI and waist circumference, which are easily measured, give an indication of the degree of EATV expansion, as does the VAT area, which is its closest approximation 23,39 . Up to 40% of the variation in EATV could be explained by various anthropometrics in 1,400 individuals representing a cross-section of the population in our recent work 24 .…”

“…Another potential limitation would be that our EAT data was automatically generated, with manual quality check only of segmentations agged as potentially awed by the fully automated model. However, our automatic quality check showed very good performance in our evaluation, comprising totally 1,800 CT-examinations (1,400 cases used in the validation process of segmentation quality 24 and 400 used to test the performance in the actual cohort) acquired with identical scanners and technical protocols. We have no reason to believe, that the performance would be inferior to the performance previously seen in images with identical technical parameters, especially since the manual quality check of the agged segmentations showed small errors in the majority of the cases, which we interpret as proof of the feasibility of the lter used.…”

“…Non-contrast cardiac CT examinations from totally 1,948 individuals were analyzed automatically for EATV and median EAT attenuation with a deep-learning based method previously developed by our group 24 .…”

“…Increased EATV has been both suggested [19][20][21] and refuted 5,22 as an independent risk factor for coronary artery disease in T2D, but the link between the conditions has not been fully elucidated. The strong co-variation between EATV and other anthropometric factors, such as BMI, waist circumference and visceral adipose tissue (VAT) 23,24 is a complicating factor in discerning an independent role of EATV in coronary pathophysiology.…”

“…33 in a study focusing on cardiac events (n = 2,068) 34 little has been published on automatic measurement of EAT parameters in larger cohorts. We have developed a fully automatic model with excellent metrics 24 for noncontrast enhanced cardiac CT images. The model is capable of segmenting the entire EATV in both noisy and/or incomplete images to speci cally target large cohort studies at population level.…”