Automated Segmentation And Clinical Target Volumes Definition For Nasopharyngeal Cancer Radiation Treatment Using Computed Tomography, Non New Net And Landmarks

Date
2023-08
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract

Purpose: Delineation of Clinical Target Volumes (CTVs) for nasopharyngeal cancer radiotherapy is inherently complex, primarily due to the variability in disease location and extent. This study introduces an auto-contouring tool developed following the NRG-HN001 protocol, demonstrating significant reconfigurability to align with the RTOG-0225 protocol. Methods: The study employed no-new-net (nnU-Net) architecture to auto-contour 27 crucial normal structures in the head and neck region, a vital step for defining CTVs according to the NRG HN001 protocol. We defined high-risk (CTV1, CTV2) and low- risk (CTV3, CTV4) CTVs using adjustable morphological geometric landmarks, thereby emulating physicians’ interpretation of protocol rules. This flexibility enabled our tool to switch between the NRG-HN001 and RTOG-0225 protocols swiftly. The results underwent evaluation using the Dice-Similarity-Coefficient (DSC) and Mean-Surface-Distance (MSD) metrics, alongside independent reviews by two physicians. Results: Comparisons between auto-contours and clinical contours for the NRG- HN001 protocol, carried out across 19 cancer patients of varying stages, revealed a DSC of 0.94±0.02, MSD of 0.4±0.4mm for high-risk CTVs, and DSC of 0.83±0.02, MSD of 2.4±0.5mm for low-risk CTVs. For the RTOG-0225 protocol, a physician reviewed 42 Na- sopharynx cases, identifying ten major deviations from the protocol. Comparisons between automatic contours and these 42 cases indicated statistically significant p-values (p=0.016) in DSC metrics, signifying the tool’s effectiveness in detecting major protocol deviations. Notably, the tool swiftly transitioned between the NRG-HN001 and RTOG-0225 protocols. We also noted statistical significance for the mean dose in the parotids and Larynx and for D1% with a p-value of 0.00 for brainstem, optic chiasm, and eyes. Conclusions: The study underscores the auto-contouring tool’s capability to delineate nasopharyngeal CTVs reliably following the NRG-HN001 protocol, and its configurability to meet the RTOG-0225 protocol requirements. The tool displayed high accuracy for nasopharyngeal CTVs and was clinically acceptable by two physicians in at least 90% of the cases. The tool’s capacity to detect significant protocol deviations and its po- tential to enhance radiation therapy process precision are noteworthy. The calculated Area Under the Curve (AUC) was found to be 0.75, indicating its good-to-moderate discriminative ability, further underscoring its clinical relevance.

Description
Keywords
nnU-net, U-net, Landmarks, Nasopharyngeal Cancer.
Citation