Detecting Coronary Artery Stenosis from Cardiac CT Images using 3D CNNs

In this paper, we propose an innovative method for fine-tuning a video pretrained model and introducing 3D representations to predict coronary artery stenosis. The input data is a collection of both contrast and non-contrast CT images. Both types of CT images are converted into 3D representations, which we call "sandwiches" of size k.We divide the sandwiches into training and validation data, and then apply 3D CNNs to detect stenosis. The experimental results are very promising, and we believe that our fine-tuned 3D CNN is practical for stenosis detection.