A fast algorithm for active contours and curvature estimation
CVGIP: Image Understanding
Efficient greedy learning of Gaussian mixture models
Neural Computation
Computer Methods and Programs in Biomedicine
| Hi-index | 0.00 |
User-based estimation of intima-media thickness (IMT) of carotid arteries leads to subjectivity in its decision support systems, while being used as a cardiovascular risk marker. During automated computer-based decision support, we had developed segmentation strategies that follow three main courses of our contributions: (a) signal processing approach combined with snakes and fuzzy K-means (CULEXsa), (b) integrated approach based on seed and line detection followed by probability based connectivity and classification (CALEXsa), and (c) morphological approach with watershed transform and fitting (WS). These grayscale segmentation algorithms yielding carotid wall boundaries has certain bias along with their own merits. We recently developed a fusion technique that was helpful in removing bias which combines two carotid wall boundaries using ground truth as an ideal marker. Here we have extended this fusion concept by taking merits of these multiple boundaries, so called, Inter-Greedy (IG) approach. Further we estimate IMT from these fused boundaries from multiple sources. Starting from the technique with the overall least system error (the snake-based one), we iteratively swapped the vertices of the profiles until we minimized its overall distance with respect to ground truth. The fusion boundary was the Inter-Greedy boundary. We used the polyline distance metric for performance evaluation and error minimization. We ran the segmentation protocol over the database of 200 carotid longitudinal B-mode ultrasound images and compared the performance of all the four techniques (CALEXia, CULEXsa, WS, IG). The mean error of Inter-Greedy technique yielded 0.32驴卤驴0.44 pixel (20.0驴卤驴27.5 µm) for the LI boundary (a 33.3%驴卤驴5.6% improvement over initial best performing technique) and 0.21驴卤驴0.34 pixel (13.1驴卤驴21.3 µm) for MA boundary (a 32.3%驴卤驴6.7% improvement). IMT measurement error for Greedy method was 0.74驴卤驴0.75 pixel (46.3驴卤驴46.9 µm), a 43.5%驴卤驴2.4% improvement.