Automatic Prostate Segmentation in Cone-Beam Computed Tomography Images using Rigid Registration* Christine Boydev ∗† , David Pasquier ‡§ , Foued Derraz ∗¶ , Laurent Peyrodie ‖ , Abdelmalik Taleb-Ahmed ¶ and Jean-Philippe Thiran †∗∗ ∗ Laboratoire d’Automatique, de M´ ecanique et d’Informatique Industrielles et Humaines (LAMIH), Universit´ e de Valenciennes et du Hainaut-Cambr´ esis, France † Signal Processing Laboratory (LTS5), ´ Ecole Polytechnique F´ ed´ erale de Lausanne, Switzerland ‡ Centre de Radioth´ erapie et d’Oncologie Galil´ ee de Lille, France § Academic Department of Radiation Oncology, Centre Oscar Lambret, Lille, France ¶ Unit´ e de Traitements de Signaux Biom´ edicaux (UTSB), Facult´ e Libre de M´ edecine, Lille, France ‖ Unit´ e de Traitements de Signaux Biom´ edicaux (UTSB), Hautes ´ Etudes d’Ing´ enieurs, Lille, France ∗∗ Department of Radiology, University Hospital Center (CHUV) and University of Lausanne (UNIL), Switzerland Email contact: christine.boydev@epfl.ch Abstract— We propose to evaluate automatic three- dimensional gray-value rigid registration (RR) methods for prostate localization on cone-beam computed tomography (CBCT) scans. In total, 103 CBCT scans of 9 prostate patients have been analyzed. Each one was registered to the planning CT scan using different methods: (a) global RR, (b) pelvis bone structure RR, (c) bone RR refined by local soft-tissue RR using the CT clinical target volume (CTV) expanded with a 1, 3, 5, 8, 10, 12, 15 or 20-mm margin. To evaluate results, a radiation oncologist was asked to manually delineate the CTV on the CBCT scans. The Dice coefficients between each automatic CBCT segmentation - derived from the transformation of the manual CT segmentation - and the manual CBCT segmentation were calculated. Global or bone CT/CBCT RR has been shown to yield insufficient results in average. Local RR with an 8- mm margin around the CTV after bone RR was found to be the best candidate for systematically significantly improving prostate localization. Index Terms— Rigid registration, segmentation, cone-beam computed tomography, image-guided radiotherapy, prostate cancer. I. I NTRODUCTION H IGHLY conformal radiation therapy has provided an opportunity for dose escalation but requires in return greater precision in treatment set-up and delivery [1]. How- ever the prostate is known to be a moving and deformable gland, and its motion nature is well-documented [2]. This limits the effectiveness of skin marks in patient set-up. Daily prostate image guidance helps to correct for inter- fractional set-up errors and to optimize tumor coverage and organs at risk avoidance. Image guidance produces images immediately before each treatment fraction, with the patient in the treatment position. These images are compared, either manually or automatically, with initial images acquired prior to treatment planning at an earlier stage. The spatial differ- ences between the planned and the daily tumor positions are referred to as patient set-up errors. *This work was financially supported by ELEKTA SAS, Boulogne Billancourt, France Techniques for daily prostate localization include kilo- voltage cone-beam computed tomography (CBCT) systems attached to the treatment unit. The CBCT image consists of a three-dimensional volumetric image resulting from the reconstruction of mutiple kV projections acquired at dif- ferent angles [3]. Prostate localization on CBCT scans is challenging due to the relatively poor image quality. It is due to low soft-tissue contrast (more scatter and hence more noise than in a CT image at the same level of radiation dose [4]) and streaking artifacts caused by possible moving gas pockets in the rectum during CBCT image acquisition (i.e. intra-fraction motion) [5]. Moreover Deurloo et al. reported that the deformation of prostate and seminal vesicles during the course of radiotherapy is small compared to organ motion and therefore in image-guided radiotherapy (IGRT) of prostate cancer, in first order, only set-up error and organ motion need to be corrected for, whereas prostate and seminal vesicles deformation can be considered as a second- order effect [2]. They concluded that this finding drastically simplifies the task of on-line image guidance for the prostate. Court et al. have developed an automatic monomodal CT/in-room-CT rigid registration of the prostate for IGRT [6]. The present study appears as a preliminary study where we propose to evaluate automatic segmentation of the prostate on CBCT scans using different methods of (multimodal) CT/CBCT rigid registration (RR) for IGRT, namely global RR, bone RR, and local soft-tissue RR focused on the prostate region expanded with several margins. To our knowledge, there are no published data about the effect of the margin size on local CT/CBCT RR quality. We also discuss and quantify the impact of air in the rectum on RR quality. II. MATERIAL AND METHODS The process of automatic CBCT prostate segmentation operates by performing RR between the (daily) CBCT and the (initial) planning CT scans. The resulting displacement is then applied to the contours manually delineated on the 35th Annual International Conference of the IEEE EMBS Osaka, Japan, 3 - 7 July, 2013 978-1-4577-0216-7/13/$26.00 ©2013 IEEE 3993