Exhibit Hall | Forum 2
Purpose: To establish a novel pipeline for daily adaptive replanning for ultra-hypofractionated prostate radiotherapy using machine learning (ML) automated planning and dose accumulation.
Methods: We retrospectively demonstrated daily adaptive replanning with dose accumulation for ten prostate patients treated on a standard linac with 4270 cGy in 7 fractions using daily iteratively-reconstructed cone-beam computed tomography (CBCT). A clinical expert contoured CTV and organs-at-risk on all CBCTs and we generated daily plans for 3 and 7 mm PTV margins using ML planning. We used a hybrid intensity/structure-based algorithm for CT-to-CBCT deformable image registration (DIR) and dose accumulation. We evaluated geometric and dosimetric DIR accuracy by comparing spatial and DVH differences between propagated and manual contours for CTV, bladder, rectum. We compared accumulated daily replanning with the reference plan using DVHs.
Results: DIR demonstrated excellent geometric accuracy with Dice similarity coefficients of 0.97±0.01, 0.99±0.01 and 0.97±0.02 for CTV, bladder and rectum. Mean distance-to-agreement for CTV, bladder and rectum were 0.3±0.3, 0.2±0.1 and 0.4±0.4 mm. Dosimetric evaluation of DIR showed DVH differences of 8±13 cGy for CTV D95%, 9±21 cGy for rectum D1cc, 33±46 cGy for rectum D35%, and 4±11 cGy for bladder D1cc (scaled to 7 fractions). We successfully executed automated daily replanning and dose accumulation across all fractions. For 3 mm margins, the reference PTV V95% was 99.7±0.1% and adapted/accumulated CTV V95% was 100±0.001%. For 3 mm margins, rectum V90% and V75% were 6.6±1.9% and 11.4±2.6% on the reference CT and 3.4±1.3% and 7.6±1.9% for adapted/accumulated plans. The same trend existed for organs-at-risk and targets for 7 mm margin.
Conclusion: In this study, we successfully demonstrated an end-to-end pipeline for daily adaptive radiotherapy using ML planning on CBCT for prostate treatment. Evaluation of the DIR demonstrated excellent spatial alignment and dosimetric accuracy. Adapted plans showed improved rectum sparing while maintaining target coverage.
Funding Support, Disclosures, and Conflict of Interest: This study is supported by the Collaborative Health Research Programs Grant, jointly funded by the Canadian Institute of Health Research (CIHR) and Natural Science and Engineering Research Council of Canada (NSERC). Chris McIntosh and Thomas Purdie receive royalties from RaySearch Inc. for machine learning based treatment planning methods.