Interactive Segmentation and Analysis of Fetal Ultrasound Images

Abstract

Abstract. The ability of ultrasound scanners to image anatomical structures in real time have led to their use in two important applications of medicine, (1) for monitoring the unborn baby (fetal ultrasound), and, (2) coronary treatment of blockages in blood vessels (intravascular ultrasound). The generated images (in the form of a continuous video) are typically noisy and contain numerous artifacts, making it difficult to isolate and measure features of interest. We explore the use of two algorithms, region growing and a variant of split/merge algorithm for segmenting sequences of fetal ultrasound images. We describe an interactive system that can rapidly process and segment an arbitrary number of features. The system exploits frame to frame coherence for accelerating the segmentation process, while at the same time combining the strengths of these algorithms and some post-processing for accurate and robust detection of features.

A toolbox for interactive segmentation based on nested partitions

“…Some lack flexibility in the interaction process [6] [7] [2] while others rely on expert knowledge from the user side [7] or make assumptions on the characteristics of the images to be segmented [3].…”