Scope
Intraoperative radiation therapy (IORT) is an intensive radiation treatment that delivers a concentrated beam of radiation to tumors as they are located during surgery. IORT allows direct radiation to the tumor while sparing normal surrounding tissue.
Treatment planning processes require exhaustive segmentation of the organs involved in treatment and those surrounding them. This requires manual segmentation of those organs, as well as tumors and security margins. In this direction, automatic and semi-automatic image segmentation techniques can significantly reduce segmentation and planning times. This activity intends to propose several multi-organ and single-organ segmentation techniques, so this process can be more easily approached, and to provide solutions for efficient interaction for automatic-segmentation manual correction.
Furthermore, during intervention or treatment imaging techniques can produce disagreeing findings, as compared to previous planning. Thus, it seems appropriate to provide means for the update according to this new knowledge. This process requires integration of registration and segmentation, as part of the development of this activity.
This segmentation task for IORT planning will be performed in the following substeps:
1.1. Multi-organ pre-labeling from manually-provided seeds using hyerarchical segmentation techniques
In this task we approach first multi-organ segmentation using hierarchical segmentation techniques, based on graph-cuts and mathematical morphology operations. The goal is being able to label from a set of seeds, a complete segmentation tree by which clinical practitioners can select the most useful segmentation level.
1.2. 3D whole-organ segmentation on minimal interaction by active contours
Greater-detail segmentation for organs of interest can complete previous labelings for those organs which require greater detail, or for internal parts of complex organs. For this task, different active contour techniques will be explored and combined with interaction techniques that allow the clinical specialist to efficiently correct automatic segmentations, by minimal actuations.
1.3. Integration of segmentation and registration techniques for intraoperative update of plannification stages
Intra-operative segmentations (in US and X-ray images) can be obtained efficiently through registration, by direct application of the registration results on the preoperative available segmentation. On the other hand, intra-operative segmentation can provide new information to the planning scenario. Both processes require proper integration between the segmentation and registration processes. In this task, techniques obtained from Activity 5 and tasks 6.1 and 6.2, will be merged, paying special attention to the performance of the final implementations.
1.4. On-field validation
Validation experiments for the developed proposals will take place in a controlled framework in order to determine accuracy, parameter influence and limitations, on synthetic tissue images (phantoms). The proposed segmentation methods will be validated in the scenarios of interest (liver surgery, interventional radiology, and intra-operative radiotherapy) using pre (usually CT and RMI) and intra-operative images. In first term, validation will be performed against manual segmentations, provided by practitioners or radiologists (objective overlap measures comparing both segmentations). In second term, easiness of use and need for interaction, will be evaluated, as to manually correct the resulting automatic segmentation resulting from independent operators.
Collaborations
Fundación Pública Andaluza para la Gestión de la Investigación en Salud de Sevilla (FISEVI)
Hospitales Universitarios Virgen del Rocío
Laboratorio de Imagen Médica, Unidad de Medicina y Cirugía Experimental, Hospital Universitario Gregorio Marañón
Biomedical Image Technologies Group, Universidad Politécnica de Madrid
Funding
Real-time multimodal medical image for complex scenarios (TEC2010-21619-C04-02)
Images
