Clinical implementation of an automatic mask generation framework for use in 2D/3D image registration for patient positioning in radiotherapy

Abstract

Clinical implementation of an automatic mask generation framework for use in 2D/3D image registration for patient positioning in radiotherapy to achieve hybrid inter- and intrafractional patient position monitoring and multi-component adaptive beam delivery based on bony anatomy

Markus Neuner1, Philipp Steininger1, Felix Sedlmayer1,2, Heinz Deutschmann1,2

1 Institute for Research and Development on Advanced Radiation Technologies (radART), Paracelsus Medical University, Strubergasse 21, 5020 Salzburg, Austria

2 University Clinic for Radiotherapy and Radio-Oncology, Paracelsus Medical University, Muellner Hauptrasse 48, 5020 Salzburg, Austria

 

The spatial alignment of pre- and intra-interventional data plays an important role in image guided radiotherapy (IGRT) to correct for day-to-day variations of the target location in the body and/or the patient on the couch, in order to guarantee an efficient treatment and to avoid adverse effects. We recently developed, evaluated and employed a (semi-)automatic intensity-based 2D/3D registration approach to correct translational and rotational displacements in the pelvic region [1-3]. It rigidly registers a computed tomography (CT) with two X-rays by maximizing the agreement in pixel intensity between the X-rays and corresponding reconstructed radiographs from the CT. In order to restrict the registration to the structures/anatomy of interest, the pixels used for alignment evaluation must be constrained to a region of interest (e.g exclude femurs when registering on the bony anatomy of the pelvis). To limit the metric evaluation to the anatomical structure of interest, we implemented an automated approach to construct regions of interest (masks) in the X-rays based on 3D segmentations from the pre-planning stage [4].

The next steps of clinical application are hybrid inter- and intra-fractional 2D/3D patient position monitoring and multi-component adaptive beam delivery based on bony anatomy, which are both in a preparatory stage. Hybrid inter- and intra-fractional 2D/3D patient position monitoring is performed during treatment and based on mega-voltage (MV) X-ray images to detect intrafractional patient movement. Image-guided setup is performed, where prior to each treatment fraction the patient’s anatomy is imaged and compared to the planned reference setup. The deviation is computed with rigid 2D/3D registration and corrected to bring the patient into alignment with the planned reference. Irradiation is started and a background patient position verification (gating/adaptation)‏ without fiducial markers (e.g. gold-seeds) is performed through MV intensity information that represents bony anatomy. The goal is to automatically detect if a patient moved between the acquired images.

Multi-component adaptive beam delivery based on bony anatomy independently registers multiple regions of interest and results in multiple rigid transformations to correct the alignment of individual anatomical structures. Indications are large target volumes where the spatial location of distinct parts correlates with nearby bony anatomy. Each split volume is then registered to the corresponding bony anatomy. Clinical cases are for instance: gynecological tumors (pelvine radiations with lymphatic system) where we can register the uterus/cervix to the lesser pelvis and the lymphatic system to the sacrum/lumbar spine, and Larynx, pharynx, tonsils, tongue base tumors where we can register to head, mandible and cervical spine.


[1] Steininger P, Neuner M, Mittendorfer C, Scherer P, Sedlmayer F, Deutschmann H. Clinical implementation of GPU-accelerated n-way 2D/3D image registration for inter-fractional patient positioning in radiotherapy; IJCARS; Vol. 6 (Supplement 1); p. S56-S57; June 2011.

[2] Warmerdam G, Steininger P, Neuner M, Sharp G, Winey B. Robustness of a 2D/3D-Image-Registration Algorithm for Cranial Image Guided Radiotherapy; submitted to MedPhys, March 2012

[3] Steininger P, Neuner M, Weichenberger H, Sharp GC, Winey B, Kametriser G, Sedlmayer F, Deutschmann H. Auto-masked 2D/3D image registration and its validation with clinical cone-beam computed tomography; submitted to Physics in Medicine and Biology; Jan 2012.

[4] Neuner M, Steininger P, Mittendorfer C, Sedlmayer F, Deutschmann H. Automatic mask generation for 2D/3D image registration with clinical images of the pelvis; IJCARS; Vol. 6 (Supplement 1); p. S54-S55; June 2011.