Abstract 13515: A Feature Classification Approach for Coronary Artery Disease Prediction Via Carotid Atherosclerosis Window
Purpose: Invasive coronary angiography is the gold standard for detecting Coronary Artery Disease (CAD). However, it is limited by cost and associated risks. There is a need for accurate non-invasive techniques for CAD diagnosis.In this paper, we have presented a computer aided diagnostic technique for CAD prediction based on features of longitudinal ultrasound carotid artery images.
Methods and Results: This technique utilizes the following feature sets: (1) gray scale features that quantitatively characterize the textural changes in the carotid far wall Intima-Media Thickness (cIMT) region along the common carotid artery that was automatically segmented; (2) cIMT value; (3) cIMT wall variability along the common tract (cIMTV). The images used in this study came from 501 consecutive patients who underwent carotid-ultrasound and coronary angiography. We used the SYNTAX Score, which was determined by experienced interventional cardiologists from the angiograms, to assess whether the cases were normal or CAD affected.
The proposed system consists of an offline training system and an online real-time system. In the offline system, the automatically extracted features were used to develop a Support Vector Machine (SVM) classifier. The on-line system applied the training parameters on the images to predict the presence of absence of CAD. The table reports the prediction performance of the SVM classifier when only the gray scale features (set(1)) and the cIMT are used (first row), when set (1) plus the cIMTV is used (second row), and when set (1) is combined with cIMT and cIMTV (bottom row). When grayscale features, cIMT, and cIMTV were all combined together and fed to the classifier, we could achieve 100% accuracy in detecting CAD.
Conclusion: The presented technique is objective, non-invasive, affordable, and the preliminary results indicate the possibility of using this technique as an adjunct diagnostic tool in a clinical setting for reliable CAD diagnosis.
- © 2013 by American Heart Association, Inc.