Cardiac Detection in Non-Contrast CT and Application to Calcium Scoring

Coronary artery disease, one of the major causes of death in Japan, is said to be related to coronary artery calcification. However, health care workers are essential to examine only the heart when investigating coronary artery disease. Also, it is very difficult to extract only calcification of the heart from Non-Contrast CT. To our knowledge, 3D cardiac extraction in non-enhanced CT cardiac images has not been performed in previous studies. In addition, the estimation of calcification score from heart extraction has not been done in previous studies. Therefore, we performed calcium score classification from cardiac CT using 3D-ResNet, which is a 3D extension of ResNet, which is highly accurate in image classification. Therefore, in this study, we propose a cardiac detection method using deep learning to improve the accuracy of calcium scoring. We conducted experiments using a dataset for cardiac detection from chest CT images and a dataset for calcium scoring from chest CT scans. The results show that the proposed method improves the accuracy of calcium scoring compared to the method without cardiac extraction. As a result, we proposed a state-of-the-art highly accurate heart detection method that has never existed before.