# Clinical feasibility of atlas‐based auto‐segmentation for organ‐at‐risk contouring in head‐and‐neck radiotherapy

## TL;DR

This study shows that automated contouring tools can produce consistent and accurate organ contours for head-and-neck radiotherapy, resulting in acceptable treatment plans.

## Contribution

The study demonstrates the clinical feasibility of atlas-based auto-segmentation for head-and-neck radiotherapy planning.

## Key findings

- Auto-generated OAR contours showed excellent agreement with manual contours in planning doses for organs farther from the target.
- Cumulative dose differences between auto-generated and manual OAR plans were within 2 Gy for 90% of patients.
- Atlas-based segmentation achieved average overlap indices and dose similarity coefficients above 76% for most OARs.

## Abstract

Head‐and‐neck (HN) contouring presents significant challenges due to the complex anatomy of the region and the proximity of organs‐at‐risk (OARs) to the target. Manual contouring is time‐consuming, labor‐intensive, and prone to inter‐ and intra‐observer variability. Additionally, contours delineated prior to treatment may not accurately reflect the patient's anatomy over the extended treatment course. This study explores the feasibility of using auto‐generated OAR contours in HN planning.

A retrospective study was conducted on 20 patients, each with a planning CT and 35 CBCT images. OARs were manually delineated and automatically generated using atlas‐based segmentation algorithms on planning CTs. Treatment plans were created via a novel two‐step optimization process, incorporating a knowledge‐based planning solution for both auto‐generated (aOAR‐plan) and manual OARs (mOAR‐plan). The accuracy of auto‐generated contours was quantified using the overlap index (OI) and dose similarity coefficient (DSC). Planning dose comparisons were performed between aOAR‐ and mOAR‐plans. Additionally, planning doses were transferred from CT to CBCTs based on clinical shifts, and contour‐based deformable registration was employed to calculate cumulative doses. Cumulative dose evaluations were performed for serial organs and parallel organs that can be fully imaged within the CBCT field.

For OARs located farther from the target, even though atlas‐based segmentation could not accurately reproduce patient anatomy, excellent agreement in planning doses was observed between the aOAR‐ and mOAR‐plans. The average OI/DSC between manual and auto‐generated contours were 85.0% ± 5.4%/87.4% ± 2.6% for the larynx, 76.0% ± 9.3%/77.0% ± 5.8% for the pharynx, 89.9% ± 4.0%/87.8% ± 2.5% for the oral cavity, 81.5% ± 10.5%/78.2% ± 5.9% and 83.2% ± 10.6%/77.8% ± 7.5% for the left and right parotid, respectively. The cumulative dose differences for OARs between aOAR‐ and mOAR‐plans were within 2 Gy for 90% of patients studied.

Automated‐contouring tools offer improvement in contour consistency, provide acceptable doses compared with manually drawn contours in HN radiotherapy.