Adaptive localized region-based fast active contour model by introducing global information in local region

To solve the problems in segmenting HIFU (High Intensity Focused Ultrasound) ultrasound image of uterine fibroids, we propose an adaptive localized region-based fast active contour model by introducing HIFU image's global information in local region, which is more accurate as well as more efficient. The proposed segmenting model incorporates HIFU ultrasound image's global information in local region to form a locally global force. Meanwhile the gray level distribution uniformity around each pixel point on the evolution curve is calculated to dynamically determine the various application condition of HIFU image's global information in local region and the shape constrained information of the uterine fibroids in HIFU images, which is assigned to overcome the sensitivity of the initialized contour by applying the locally global force when segmenting HIFU ultrasound image of uterine fibroids. By using the calculated gray level distribution uniformity around each pixel point on the evolution curve, the adaptive localized region-based fast active contour model adaptively changes the local radius of the localized region, and then dynamically adjusts the size of localized region during the evolution process of the active contour curve, achieving more accurate and more efficient segmentation results of HIFU ultrasound image of uterine fibroids. By applying the same localized region to calculate the local forces of adjacent pixel points on the evolution curve, our method further improves the segmentation efficiency, finally achieving accurate and efficient segmentation of HIFU ultrasound image of uterine fibroids. The experimental results show that compared with recently proposed MSLCV (Multi-scale and Shape Constrained Localized C-V) model, our method solves the problems in segmenting HIFU ultrasound image of uterine fibroids as well as improves the segmentation accuracy and increases the average segmentation efficient by 84.6%. © 2016, Science Press. All right reserved.