ESTRO 2024 - Abstract Book

S3789

Physics - Image acquisition and processing

ESTRO 2024

The random walker (RW) algorithm is a state-of-art and widely accepted method for image segmentation. After selecting two interactive labels, one as the “target seed” and the other as the “background seed”, the RW is released for each pixel in the image and the probability is computed that each pixel’s RW will first arrive at either of the two seeds. After all, the weighted graph is modeled, in which each pixel corresponds to one of the seeds which is connected to neighboring pixels by edges. That is, the edge weights are assigned based on the similarity to the respective seed, derived from differences in image intensity. In our study, the RW algorithm was used to segment the sternum in thorax DTS images, which were imaged using different acquisition rotation trajectories. The accuracy of the RW segmentation reflected the quality of the DTS images.

Results:

For the purpose of enhancing the edges of the sternum in thorax case, sagittal DTS images were acquired, which were imaged along the gantry angle at 270°. The intended DTS image acquisition rotation trajectory included: (1) using dual 180° arc for full-projection scan (full-AB); (2) using dual 100° arc for half-projection scan (half-AB); (3) using single 100° arc for half-projection scan (half-A); (4) using single 85° arc for partial-projection scan (partial-A). The actual image acquisition angles were determined by subtracting both the initial 20° and the final 20° rotation angles, accounting for the buffer time required for the gantry to achieve a consistent rotation speed and to come to a complete stop. In figure 1, the RW segmentation process is illustrated for segmenting the body of sternum, including the xiphoid process: (A) shows the sagittal DTS image with target seed (green) and background seed (blue); (B) displays the RW output mask; (C) represents the weighted graph; (D) shows the target outline in red.

In figure 2, the segmentation results achieved by RW algorithm are presented: (A), (B), (C) and (D) display the segmentation results of the sternum body and xiphoid process in full-AB, half-AB, half-A and partial-A DTS sagittal images, respectively. (E) and (F) show the segmentation results of sternum, including the manubrium, body and xiphoid process, in half-A and partial-A DTS sagittal images, respectively. (E#) and (F#) represent the histogram of the grayscale values within the sternum outline in half-A and partial-A DTS sagittal images, respectively.