Rosebrugh Bldg, Toronto, ON M5S 3G9
Room: RS 211
One of the largest uncertainties in radiotherapy treatment planning for head and neck cancer is contouring the gross tumor volume. Under-contouring or over-contouring the tumor can increase the risk of negative health outcomes for the patient. Currently, endoscopic exams are only used in clinic to assess the tumor but the use of endoscopic exams in treatment planning is expected to improve contouring accuracy. Registration of the endoscopic images to the CT image allows information from the endoscopic images to be spatially mapped onto the CT. After registration, tumor margins viewed under endoscopy can be directly contoured on endoscopic images and projected onto the CT used in treatment planning.
Develop an automatic endoscopy to CT registration procedure using electromagnetic (EM) sensors and image based registration (IBR) methods to minimize registration error. Inputs to the registration pipeline are endoscopic images, EM tracker positions and the CT image. The CT image is used to render a virtual endoscopic image at the EM tracker position. Both real and virtual images are fed into an optimization routine to find the best virtual image used to contour the tumor. The registration was quantified using the feature registration error (FRE) measured between pairs of features picked in the real and virtual endoscopic images and the contour reprojection error (CRE) measured between the endoscopy-projected contour and the CT contour.
On 6 images from 2 patient cases the FRE was 2.3mm – 4.4mm after IBR. On 2 endoscopic contours on a rigid phantom the CRE was 1.9-14.5 mm.
An endoscopy to CT registration pipeline has been developed for radiotherapy treatment planning and achieves errors smaller than the 5mm clinical uncertainty margin. More clinical data and further research into IBR metrics is required to improve registration results.