Margin Reduction Based on Comparison of Histopathology With Delineations on CT, MRI, and FDG-PET in Laryngeal and Hypopharyngeal Cancer

Abstract

Tumor delineation is one of the weakest links in the radiation therapy (RT) treatment chain. The margin needed to determine the clinical target volume (CTV) is also questionable. In order to improve treatment, more insight into the errors made by delineation is needed. In this study, gross tumor volume (GTV) delineations on computed tomography (CT), magnetic resonance imaging (MRI), and the automatic segmentation on 18fluorodeoxyglucose positron emission tomography (FDG-PET) of laryngeal and hypopharyngeal squamous cell carcinoma are validated on histopathology. Twenty-one patients were included with a laryngeal (n=14) or hypopharyngeal (n=7) tumor (T3/T4). Before total laryngectomy/partial pharyngectomy, CT, MRI (T1weighted and T2-weighted), and FDG-PET images were made in RT positioning mask. The GTV was delineated in consensus by 3 experienced observers on CT and MRI. FDG-PET was delineated semi-automatically using a Gaussian mixed model based threshold. Tumor has been delineated by a pathologist on whole-mount hematoxylin and eosin stained (HE) histopathologic sections (intersections of approximately 3 mm) and afterwards digitized and reconstructed to a 3 dimensional specimen. This 3D specimen was then registered to the imaging prior to laryngectomy. The GTVs were compared with the histopathologic-based tumor delineation in 3 dimensions, including the microscopic extension. The margin around the GTV needed to ensure 95% coverage of the histopathologically determined tumor surface, was 3 mm (SD 2), 4 mm (SD 2) and 4 mm (SD 2) for CT, MRI and FDG-PET, respectively. In the semi-automatic delineation of FDG-PET, for 1 case a margin of 9 mm was needed, probably due to large necrotic parts in the tumor. The part of the tumor included in the GTV (sensitivity) amounted to 86% (SD 9), 80% (SD 11) and 79% (SD 11) for CT, MRI and FDG-PET, respectively, while the overestimation of the tumor volume amounted to 87% (SD 45), 62% (SD 33) and 50% (SD38). The positive predictive value (PPV), on CT, MRI, and FDG-PET was found to be 53% (SD 14), 59% (SD 13), and 66% (SD 15), respectively. In order to cover 95% of the histopathologic tumor surface, including the HE-determined microscopic extension, the margin added to the GTVs needed to be 6-7 mm (average margin + 1.5 times SD). This margin indicates that the standard clinical target volume (CTV) margin of 10 mm could be reduced. FDG-PET has the highest PPV; therefore semi-automatic delineation of FDG-PET can be used as an initial delineation of the GTV. This delineation must be reviewed by a physician and, if necessary, modified based on additional information from high resolution CT or MRI.