|M.Sc Student||Toledano Michal|
|Subject||A Learning System for Detection and Characterization|
of Coronary Arteries Stenosis Using Multi-Detector
|Department||Department of Biomedical Engineering||Supervisors||Professor Michael Lindenbaum|
|Professor Rafael Beyar|
Aim: To develop and evaluate an artificial intelligence (AI) based system for detection of coronary artery stenoses using multi-detector computerized tomography coronary angiography (CTCA).
Methods: The learning system, based on a well known methodology (SVM), developed for automatic detection of stenoses > 30% , was trained using CTCA reformation images of 16 patients with coronary artery disease, proved previously by standard invasive coronary angiography (ICA). The CTCA studies were performed on a 16-slice scanner (Brilliance - Philips Medical Systems) with coronary stenoses detected and quantified by a team of expert physicians.
The system was tested on 42 coronary artery segments and the results were compared with human interpretation of the CTCA and with the Quantitative Coronary Angiography (QCA) of the ICA.
Results: The sensitivity, specificity, positive and negative predictive values of the automated detection system of coronary artery stenoses, compared with the human interpretation of the CTCA were 96.4% (27/28), 92.9% (13/14), 96.4% (27/28), 92.9% (13/14) respectively. When comparing to the QCA, the sensitivity, specificity, positive and negative predictive values were 91.7% (22/24), 66.7% (12/18), 78.6% (22/28), 85.7% (12/14) respectively, which were very similar to the sensitivity, specificity, positive and negative predictive values obtained from the comparison of the human interpretation of the CTCA with QCA: 95.8% (23/24), 72.2% (13/18), 82.1% (23/28), 92.9% (13/14) respectively.
Conclusions: Our results demonstrate that automatic detection of coronary artery stenoses from CTCA reformations images based on an artificial intelligence algorithm, is feasible and comparable with the human interpretation of computed tomography.