In this work, the development of an automatic system for quantifying the coronary artery calcium (CAC) score in the context of both non-contrast and contrast CT scans based on the use of the 2D-UNet slice-by-slice segmentation network for the following tasks is described: (1) localization of the heart region of interest, (2) delineation of the coronary arteries within the scanned body regions of interest, as well as (3) the detection of calcium within the delineated coronary arteries based on the score range of 130-199 to 400+. In accordance with the Agatston score computation method for calcium quantification, the system quantifies weighted calcium scores that are further grouped within predefined established risk categories (0, 1-100, 101-300, > 300) with the use of the Grad-CAM visualization method. The new method introduced here has the ability to estimate the burden of the coronary arteries in patients in relation to myocardial infarction risk by identifying those with scores above 300 as individuals with high cardiovascular risk. Five seconds or less can be used for the processing of either gated chest scans as well as for the average scanning protocols of the body without significantly affecting the sensitivity of the system relative to the manual scoring of the coronary artery calcium in the CT scan. Moreover, the system has the ability to provide uniform diagnostic sensitivity with DICE indexes of above 0.85. There is excellent correlation as well as the ability to reduce inter-observer variability associated with the manual computation of the CAC scoring method. The new system has the capability for transparency for the treating professionals in terms of providing them with the capability of comprehending the diagnostic outcomes via the use of the AI method without the need for additional imaging as well as not requiring the use of additional equipment.
M et al. (Thu,) studied this question.