Technical note: Air bubble‐induced performance degradation in automatic rectum segmentation from cone‐beam CT. Issue 3 (25th January 2022)

• Record Type: Journal Article
• Authors: Zhou, Hanyue; Cao, Minsong; Ma, Martin; Yoon, Stephanie; Kishan, Amar; Ruan, Dan
• Is Part Of: Medical physics. Volume 49:Issue 3(2022)
• Page Start: 1754
• Page End: 1758
• Publication Date: 2022-01-25
• Subjects: air bubbles -- automatic rectum segmentation -- cone‐beam CT -- Feldkamp artifacts -- mixed linear regression; Medical physics -- Periodicals; Medical physics; Geneeskunde; Natuurkunde; Toepassingen; Biophysics; Periodicals; Periodicals; Electronic journals; 610.153
• DOI: 10.1002/mp.15443
• Languages: English
• ISSNs: 0094-2405

Abstract

Abstract: Purpose: Cone‐beam computed tomography (CBCT) is widely used for daily anatomy monitoring and can be a potential source to support adaptation. However, low image quality and artifacts limit CBCT's clinical utility. Peristalsis and air bubbles can cause severe artifacts in pelvic CBCT. We have observed that severe air bubble‐induced Feldkamp artifacts in the rectum may contribute to low automatic segmentation accuracy. Materials and Methods: In this study, air bubbles within the rectum were extracted and automatic rectum segmentation performance was measured in Dice similarity coefficient (DSC). A Gaussian mixture model (GMM) was used to characterize their correlation, and an expectation‐maximization (EM) approach was used to solve the corresponding parameter estimation and decouple the impact from air bubbles versus other image attributes based on cluster memberships. Postprostatectomy patient data with high variability in air bubble size and shape were used in this study to reveal the regression relationship. Results: GMM identified two distinct correlative relations between the air‐bubble severity in the rectum and the rectum prediction DSC: one showed strong negative dependency of segmentation performance on air bubble presence, and the other one had mild‐to‐moderate dependency that suggested another group of contributing factors influencing rectum segmentation, such as the inconsistent presence of fiducial seeds and shape extremes. Conclusion: The presence of severe air bubbles contributes semilinearly to performance degradation in automatic rectum segmentation. A good correction mechanism may boost the accuracy and consistency of pelvic segmentation.