Automatic delineation of body contours on cone-beam CT images using a delineation booster

In radiotherapy, cone-beam computerized tomography (CBCT) scans are used for position correction for various tumour sites. At the start of the treatment, a CT scan that serves as input for a treatment planning is acquired. A CBCT scan is made prior to the irradiation of the tumour. Because there might be significant interfractional tumour movement, online recalculation of the dose improves decision making on how to proceed. A prerequisite for such recalculation is an accurately delineated body contour. In this note, we present an automatic delineation method for the body contour in the unprocessed CBCT scans, that employs a novel delineation boosting technique. The main idea of this technique is to construct an accurate delineation by combining the strength of several edge detectors in an innovative way. Quantitative validation reveals that the algorithm performs comparably with the manual delineations of two trained observers. Furthermore, because of the generic nature of the delineation boosting procedure, the algorithm can easily be extended with additional edge detectors to further increase the accuracy. Finally, the processing time of one scan when delineated manually is 3 h, and the total processing time is 24 min for one scan if the algorithm is used in its present form. Current investigation includes the conversion of the Matlab algorithm to C++ and the development of a visual tool to quickly detect which automatically delineated slices need manual correction. From this we expect further speeding up of the process, allowing online computation.