Video-based positioning control in a treatment room

The various professionals (oncologists, medical physicists, radiation therapist, etc.) involved in the treatment of cancer by radiation therapy express more and more acutely a real need to have a global view of the treatment system in order to monitor the patient and all the equipment in the room - be it mobile (or even robotized) or not - to improve patient safety. This global view would help reduce the number of human interventions, errors in physical interaction (involuntary movements, collisions, etc.), and errors in the global configuration for the patient. To answer this need, this PhD manuscript presents a video-surveillance module using a numerical model of the elements in the scene (patient, couch, irradiation arm, etc.) to monitor the movement of these elements in order to avoid any event that could put the patient, staff or equipment in danger. Monitoring the treatment room means checking, at any moment, that its current configuration is consistent with what was planned upstream (in the Treatment Planning System [TPS]) and with the standards of radiotherapy, and if need be to, to inform the medical staff, so that correcting actions may be undertaken. We have developed a generic model which allows the creation of numerical models from the data contained in the patient record (in the TPS), enriched with other visual, geometric and 'semantic' information. All the numerical models used to create a virtual environment correspond to the scene as seen by video cameras. Location and orientation of the elements are calculated by 3D registration, a choice justified by the characteristics of radiation treatment rooms (controlled environment, low dynamics, highly constrained construction specifications, very similar treatment rooms, etc.). The (geometric, physical and mechanical) information items in the model allow to reduce the overall complexity of the algorithms. The method for motion tracking we have implemented exploits the characteristics and constraints of the elements included in the numerical model, specifically the 'descriptors' which correspond to the 3D modeling of primitives used to track the element in the video stream. We also present a dissimilarity function that we have defined, resulting from the merging of two well-known dissimilarity functions, respectively based on the chamfer distance and the notion of non-overlap. A fuzzy formalism improves the robustness of the method against the noise generated by acquisition and processing. When the treatment session takes place, the module evaluates the current configuration (positions acquired for elements, using the tracking module and global parameters of the scene), to interact with medical staff in relation to what has been defined in pretreatment. We present results of tests performed in a radiotherapy room but in the absence of irradiation and patient (the patient was either replaced by a phantom positioned on the couch or by one researcher playing his/her role). These results demonstrate the feasibility and efficiency of the surveillance module. (author)