Deformable registration-based segmentation of the bowel on Megavoltage CT during pelvic radiotherapy

• Published in: Physica medica Vol. 32; no. 7; pp. 898 - 904
• Main Authors: Perna, L., Sini, C., Cozzarini, C., Agnello, G., Cattaneo, G.M., Hysing, L.B., Muren, L.P., Fiorino, C., Calandrino, R.
• Format: Journal Article
• Language: English
• Published: Italy Elsevier Ltd 01.07.2016
• Subjects: Bowel motion; Bowel segmentation; Deformable registration; Feasibility Studies; Humans; Image Processing, Computer-Assisted - methods; Intestines - diagnostic imaging; Male; Observer Variation; Pelvis - radiation effects; Prostatic Neoplasms - diagnostic imaging; Prostatic Neoplasms - radiotherapy; Radiology; Radiotherapy Planning, Computer-Assisted; Time Factors; Tomography, X-Ray Computed; Bowel segmentation; Deformable registration; Bowel motion
• ISSN: 1120-1797; 1724-191X
• DOI: 10.1016/j.ejmp.2016.06.009

•Overview of bowel motion in pelvic radiotherapy.•Evaluation of the potential of daily MVCT to visualize the bowel.•Evaluation of the deformable registration and contour propagation methods for bowel.•Evaluation of the time sparing for bowel segmentation compared to manual contouring.•Evaluation of the impact on bowel DVH of the deformable registration based method. During pelvic radiotherapy bowel loops (BL) are subject to inter-fraction changes. MVCT images have the potential to provide daily bowel segmentation. We assess the feasibility of deformable registration and contour propagation in replacing manual BL segmentation on MVCT. Four observers delineated BL on the planning kVCT and on one therapy MVCT in eight patients. Inter-observer variations in BLs contouring were quantified using DICE index. BLs were then automatically propagated onto MVCT by a commercial software for image deformation and subsequently manually corrected. The agreement between propagated BL/propagated+manually corrected BL vs manual were quantified using the DICE. Contouring times were also compared. The impact on DVH of using the deformable-registration method was assessed. The same procedures were repeated on high-resolution planning-kVCT and therapy-kVCT. MVCTs are adequate to visualize BL (average DICE: 0.815), although worse than kVCT (average DICE:0.889). When comparing propagated vs manual BL, a poor agreement was found (average DICE: 0.564/0.646 for MVCT/KVCT). After manual correction, average DICE indexes increased to 0.810/0.897. The contouring time was reduced to 15min with the semi-automatic approach from 30min with manual contouring. DVH parameters of propagated BL were significantly different from manual BL (p<0.0001); after manual correction, no significant differences were seen. MVCT are suitable for BL visualization. The use of a software to segment BL on MVCT starting from BL-kVCT contours was feasible if followed by manual correction. The method resulted in a substantial reduction of contouring time without detrimental effect on the quality of bowel segmentation and DVH estimates.