Fig. 3

Examples show potential causes of discordances of Agatston scores between chest CT and cardiac CT. For each case(A-H), the upper row is AI-based scoring on chest CT, and the lower row is the manual measurement on cardiac CT. CAC lesions were annotated with different colours (i.e., green for LM, yellow for LAD, blue for CX, red for RCA, and pink for non-coronary calcifications). A. False negatives due to motion artifact, B. Underestimation due to motion artifact (Agatston score:57.7 versus 225.0 between chest CT and cardiac CT). C. False positive due to image noise on chest CT, it is not identified because of tiny size and lower density (< threshold of 130HU) on standard cardiac CT. D. Overestimation due to motion artifact (Agatston score:126.0 versus 88.0 between chest CT and cardiac CT). E. Segmentation error, CAC of LM (green on cardiac CT) is falsely identified as calcification of LAD (yellow on chest CT). F. Segmentation error combined with motion artifact, CAC of LM (green on cardiac CT) is wrongly identified as CAC of LAD (yellow on chest CT) and blurred due to motion artifact. G. Segmentation error, calcification of liver is misidentified as CAC of RCA. H. Segmentation error, part of CAC of LAD is misidentified as non-coronary calcification, resulting in underestimation (Agatston score:48.1 versus 713.2 between chest CT and cardiac CT)