Daily dose monitoring with atlas-based auto-segmentation on diagnostic quality CT for prostate cancer

To evaluate the feasibility of daily dose monitoring using a patient specific atlas-based autosegmentation method on diagnostic quality verification images.

Seven patients, who were treated for prostate cancer with intensity modulated radiotherapy under daily imaging guidance of a CT-on-rails system, were selected for this study. The prostate, rectum, and bladder were manually contoured on the first six and last seven sets of daily verification images. For each patient, three patient specific atlases were constructed using manual contours from planning CT alone (1-image atlas), planning CT plus first three verification CTs (4-image atlas), and planning CT plus first six verification CTs (7-image atlas). These atlases were subsequently applied to the last seven verification image sets of the same patient to generate the auto-contours. Daily dose was calculated by applying the original treatment plans to the daily beam isocenters. The autocontours and manual contours were compared geometrically using the dice similarity coefficient (DSC), and dosimetrically using the dose to 99% of the prostate CTV (D99) and the D5 of rectum and bladder.

The DSC of the autocontours obtained with the 4-image atlases were 87.0% ± 3.3%, 84.7% ± 8.6%, and 93.6% ± 4.3% for the prostate, rectum, and bladder, respectively. These indices were higher than those from the 1-image atlases (p < 0.01) and comparable to those from the 7-image atlases (p > 0.05). Daily prostate D99 of the autocontours was comparable to those of the manual contours (p = 0.55). For the bladder and rectum, the daily D5 were 95.5% ± 5.9% and 99.1% ± 2.6% of the planned D5 for the autocontours compared to 95.3% ± 6.7% (p = 0.58) and 99.8% ± 2.3% (p < 0.01) for the manual contours.

With patient specific 4-image atlases, atlas-based autosegmentation can adequately facilitate daily dose monitoring for prostate cancer.