Weixin Si, Presenter: Nothing to Disclose

Weiming Wang PhD, Abstract Co-Author: Nothing to Disclose

Zhiyong Yuan PhD, Abstract Co-Author: Nothing to Disclose

Pheng Ann Heng PhD, Abstract Co-Author: Nothing to Disclose

In order to improve the performance of ultrasound-guided high intensity focused ultrasound (HIFU) therapy for uterine fibroid ablation, we develop a fast and accurate approach for uterine fibroid segmentation in HIFU images.

Ultrasound-guided HIFU therapy is a new type of noninvasive procedure for uterine fibroid ablation and it is desirable to design a reliable method to segment the uterine fibroids from HIFU images, which usually exhibit intensity inhomogeneity and blurred edges. To this end, we propose to segment HIFU images by combining local region information and shape constraint. Specifically, we employ a localized active contour model to deal with inhomogeneous images so that the contour can precisely capture the boundaries in nonuniform regions. In order to avoid boundary leakages at blurred edges, we incorporate a priori shape information to constrain the contour during the evolution. Lastly, we adopt the multiscale approach to accelerate the performance.

We test the proposed approach with twenty HIFU images, and the size of each image is 524*413. To validate the performance of our approach, we compare it with other five classical segmentation methods: GAC (Geodesic Active Contours) model, CV (piecewise constant) model, LCV (piecewise smooth) model, RSF (Region Scalable Fitting) model and LGF (Local Gaussian Fitting) model. Experimental results show that our approach outperforms all the other methods and can accurately segment the uterine fibroids from HIFU images. We further employ the widely used Dice Similarity Coefficient (DSC) measure to quantitatively evaluate the segmentation methods. In all the tested images, our approach achieves the highest accuracy of 0.94 in DSC measurement. Lastly, the segmentation time of our approach is about 15 seconds, which is only one eighth of the LCV model.

We propose a fast and accurate approach to segment uterine fibroids from HIFU images. The approach can automatically detect the boundaries of uterine fibroids after the user specifies the initial contour. In all the tested images, the segmentation results satisfy the requirements of the doctors, indicating its potential in ultrasound-guided HIFU therapy.

The operation time of ultrasound-guided HIFU therapy is very long, and it is thus essential and meaningful to design a fast and accurate method for uterine fibroid segmentation in HIFU images.

Si, W, Wang, W, Yuan, Z, Heng, P, Shape Constrained Active Contour Model for High Intensity Focused Ultrasound Image Segmentation.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14002832.html