Click here to

Session: Multi-Disciplinary General ePoster Viewing [Return to Session]

Evaluation of a Deep-Learning Auto-Segmentation Model Trained with Planning CT On Regular CBCT and ICBCT Images for Prostate Online Adaptive Therapy

J Duan1*, X Feng2, J Latorre1, Q Chen1, (1) university of kentucky, Lexington, KY, (2) Carina Medical LLC, Lexington, KY


PO-GePV-M-125 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: Cone-beam Computed tomography (CBCT) based treatment planning is an attractive solution for online adaptive therapy (ART). A commercial deep-learning auto-segmentation (DLAS) solution has been developed to provide accurate segmentation on iterative CBCT (iCBCT). However, this solution is limited to the Varian Ethos platform. The purpose of this study is to evaluate the performance of a third-party DLAS program trained with planning CT only on regular CBCT as well as iCBCT outside of Ethos platform.

Methods: A total of 43 intact prostate CBCTs or iCBCTs were retrospectively selected from Varian Truebeam and Halcyon system. Organs-at-risk (OARs) were manually segmented on the CBCT/iCBCT. A third-party commercial DLAS software trained on planning CT was used to segment all cases. Geometric agreements between DLAS and manual were computed. The prostate was reviewed and edited to create a PTV structure. Varian’s Rapid-plan was used to create 2-arc VMAT plans using un-modified DLAS OAR contours (plan-DLAS) vs using manual OAR contours(plan-manual). Both plans use the same prescription dose (70Gy). For both plans, the dosimetric metrics evaluated on the same manual structures following the RTOG-0815.

Results: The DLAS achieved similar quality on CBCT and iCBCT. The average Dice achieved by DLAS for bladder, rectum, left and right femoral head were 0.88±0.08, 0.82±0.07, 0.93±0.09, and 0.92±0.09 respectively. All plans are acceptable under RTOG-0815. PlanDLAS has statistically significant increase in bladder D25 by 0.39Gy than plan-manual on iCBCT. PlanDLAS on standard CBCT produces statistically significant dose increase for bladder (0.45 Gy, 0.38 Gy, and 0.33 Gy for D35, D25, and D15) and rectum (0.50 Gy for D50).

Conclusion: The DLAS, even if only trained with planning CT, can produce good quality OAR contouring on both regular CBCT and iCBCT for prostate region. Our results suggested that the DLAS OAR contour requires minimal modification for prostate online adaptive therapy.

Funding Support, Disclosures, and Conflict of Interest: NIH:R44CA254844 NIH:75N91020C00048 Varian Research grant: 75N91020C00048 Xue Feng is employee of Carina Medical LLC. Quan Chen is shareholder of Carina Medical LLC


Not Applicable / None Entered.


IM/TH- image Segmentation: General (Most aspects)

Contact Email