Click here to

Session: Therapy General ePoster Viewing [Return to Session]

Are Auto-Generated Organ-At-Risk Contours Good Enough in Head-And-Neck Radiotherapy?

H Liu1*, B Sintay2, D Wiant3, (1) Cone Health Cancer Center, Kernersville, NC, (2) Cone Health, Greensboro, NC, (3) Cone Health Cancer Center, Greensboro, NC


PO-GePV-T-279 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: Manual contouring for head-and-neck (HN) radiotherapy is a time- and labor-intensive process, subject to inter- and intra-observer variabilities. Furthermore, contours drawn prior to treatment do not necessarily represent the patient anatomy over an extended treatment course. This study investigates the feasibility of using auto-generated organ-at-risk (OAR) contours for planning.

Methods: 20 patients (each with a planning CT and 35 CBCT images) were retrospectively studied. All OARs were both manually delineated, and automatically contoured with atlas-based segmentation algorithms on planning CTs. Treatment plans were generated via a novel two-step optimization process accessing knowledge-based planning to create plans on the auto-generated OARs (aOAR-plan) and manual OARs (mOAR-plan). Overlap index (OI) and dose similarity coefficient (DSC) were used to quantify auto-generated contour accuracy. Plans comparisons were performed between aOAR-plans and mOAR-plans for all OARs. Plannng doses were transferred from CT to CBCTs based on clinical shifts, and contour-based deformable registrations were used to obtain cumulative doses. The cumulative dose evaluations were performed for serial organs (mandible, brainstem and cord), and parallel organs that could be fully imaged on CBCT (parotid glands, oral cavity, larynx and pharynx).

Results: For OARs (optical structures, etc.) far away from the target, even though atlas-based segmentation did not exactly reproduce patient anatomy, very good agreement of planning doses between aOAR-plans and mOAR-plans were observed. The average OI/DSC between manual and auto-generated contours were 85.0%±5.4%/87.4%±2.6% for larynx, 76.0%±9.3%/77.0%±5.8% for pharynx, 89.9%±4.0%/87.8%±2.5% for oral cavity, 81.5%±10.5%/78.2%±5.9% and 83.2%±10.6%/77.8%±7.5% for left and right parotid, respectively. The cumulative dose difference (maximum dose for mandible, brainstem and cord; mean dose for larynx, pharynx, oral cavity, left and right parotid) between aOAR-plan and mOAR-plans are within 2 Gy for 90% of patients studied.

Conclusion: Automated-contouring tools offer improvement in contour consistency, and provide acceptable doses compared with precision contours drawn by humans.

Funding Support, Disclosures, and Conflict of Interest: This research was supported by a grant from Varian Medical Systems, Palo Alto, CA.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email