Machine learning approach to cardiac CT structure extraction

• Main Author: Wan, Zi Qing
• Other Authors: Lin Zhiping
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/149317
• Institution: Nanyang Technological University

Coronary artery centerline extraction is an important and challenging prerequisite for coronary artery stenosis and atherosclerosis evaluation. Deep learning has recently been demonstrated to be able to process medical images effectively. However, for coronary artery centerlines, it is still challenging because of the small and complex structure and the noisy image data. This Final Year Project aims to use the datasets which have fewer annotation to train a Coronary Neural Network (CorNN) to track the coronary artery centerline direction and then combine the CorNN tracking and another segmentation network such as U-Net simultaneously for multi-tasking. In this work, each medical image is divided into many small patches. After pre-processing, the patches are used for training. Our experiment firstly trains the CorNN using a fully labeled dataset with both radius and direction information. After that, another partially labeled dataset with direction information only is added to increase the number of training patches. Then the centerline network and the segmentation network are combined as one network and trained. The combined network can predict coronary artery centerline and segmentation simultaneously. Results show the encouraging improvement of coronary artery centerline extraction by the proposed methods.