Cascaded deep learning‐based auto‐segmentation for head and neck cancer patients: Organs at risk on T2‐weighted magnetic resonance imaging

Abstract

To investigate multiple deep learning methods for automated segmentation (auto-segmentation) of the parotid glands, submandibular glands, and level II and level III lymph nodes on magnetic resonance imaging (MRI).Outlining radiosensitive organs on images used to assist radiation therapy (radiotherapy) of patients with head and neck cancer (HNC) is a time-consuming task, in which variability between observers may directly impact on patient treatment outcomes. Auto-segmentation on computed tomography imaging has been shown to result in significant time reductions and more consistent outlines of the organs at risk. Methods: Three convolutional neural network (CNN)-based auto-segmentation architectures were developed using manual segmentations and T2-weighted MRI images provided from the American Association of Physicists in Medicine (AAPM) radiotherapy MRI auto-contouring (RT-MAC) challenge dataset (n = 31). Auto-segmentation performance was evaluated with segmentation similarity and surface distance metrics on the RT-MAC dataset with institutional manual segmentations (n = 10). The generalizability of the autosegmentation methods was assessed on an institutional MRI dataset (n = 10). Results: Auto-segmentation performance on the RT-MAC images with institutional segmentations was higher than previously reported MRI methods for the parotid glands (Dice: 0.860 ± 0.067, mean surface distance [MSD]: 1.33 ± 0.40 mm) and the first report of MRI performance for submandibular glands (Dice: 0.830 ± 0.032, MSD: 1.16 ± 0.47 mm). We demonstrate that highresolution auto-segmentations with improved geometric accuracy can be generated for the parotid and submandibular glands by cascading a localizer CNN and a cropped high-resolution CNN. Improved MSDs were observed between automatic and manual segmentations of the submandibular glands when a lowresolution auto-segmentation was used as prior knowledge in the second-stage CNN. Reduced auto-segmentation performance was observed on our institutional MRI dataset when trained on external RT-MAC images; only the parotid gland auto-segmentations were considered clinically feasible for manual correction (Dice: 0.775 ± 0.105, MSD: 1.20 ± 0.60 mm). Conclusions: This work demonstrates that CNNs are a suitable method to auto-segment the parotid and submandibular glands on MRI images of patients with HNC, and that cascaded CNNs can generate high-resolution segmentations with improved geometric accuracy.Deep learning methods may be suitable for auto-segmentation of the parotid glands on T2-weighted MRI images from different scanners, but further work is required to improve the performance and

Comparative parotid gland segmentation by using ResNet‐18 and MobileNetV2 based DeepLab v3+ architectures from magnetic resonance images

“…Table 8 provides a summary of studies reporting parotid gland segmentation. When the literature is examined, 8–30 it is seen that the parotid glands are generally segmented separately as left and right. At this point, the proposed system can segment both the left and right parotid glands.…”

ContourGAN: Auto‐contouring of organs at risk in abdomen computed tomography images using generative adversarial network

“…Cascaded networks [20][21][22][23] use UNets in a nested manner and redesign the connections between the encoder and decoder layers to accumulate semantically diverse characteristic features at the decoder path, resulting in a very versatile feature fusion technique. Dense networks 24,25 concatenate the feature maps from multiple levels in the encoder-decoder architecture, which helps to accurately place the learned features from the encoder to the output stage.…”

Region of Interest focused MRI to Synthetic CT Translation using Regression and Classification Multi-task Network

“…The utility of MR imaging in radiation therapy workflow has been of interest over the past few decades for therapy monitoring [38] and for tumor depiction and delineation [2]. In recent times, the increased interest in MR-only radiation therapy workflow has led to an active research in the areas of fully automated tumor and organs-at-risk segmentation on MR to aid in therapy planning [39] [40] [41]. The electron density information required in therapy planning which is traditionally provided by a CT image is being replaced with MR derived synthetic CT (sCT) [42].…”