Deep learning‐based auto segmentation using generative adversarial network on magnetic resonance images obtained for head and neck cancer patients

Abstract

Purpose Adaptive radiotherapy requires auto‐segmentation in patients with head and neck (HN) cancer. In the current study, we propose an auto‐segmentation model using a generative adversarial network (GAN) on magnetic resonance (MR) images of HN cancer for MR‐guided radiotherapy (MRgRT). Material and methods In the current study, we used a dataset from the American Association of Physicists in Medicine MRI Auto‐Contouring (RT‐MAC) Grand Challenge 2019. Specifically, eight structures in the MR images of HN region, namely submandibular glands, lymph node level II and level III, and parotid glands, were segmented with the deep learning models using a GAN and a fully convolutional network with a U‐net. These images were compared with the clinically used atlas‐based segmentation. Results The mean Dice similarity coefficient (DSC) of the U‐net and GAN models was significantly higher than that of the atlas‐based method for all the structures (p < 0.05). Specifically, the maximum Hausdorff distance (HD) was significantly lower than that in the atlas method (p < 0.05). Comparing the 2.5D and 3D U‐nets, the 3D U‐net was superior in segmenting the organs at risk (OAR) for HN patients. The DSC was highest for 0.75–0.85, and the HD was lowest within 5.4 mm of the 2.5D GAN model in all the OARs. Conclusions In the current study, we investigated the auto‐segmentation of the OAR for HN patients using U‐net and GAN models on MR images. Our proposed model is potentially valuable for improving the efficiency of HN RT treatment planning.

Artificial Intelligence in CT and MR Imaging for Oncological Applications

“…Cross-sectional CT and MRI are an integral part of the diagnostic workup. Applications of novel narrow-specific AI tasks in these imaging techniques have shown promise for data acquisition, image segmentation and registration, and assessment of tumor responses to therapy in brain tumors [30], breast [32], head and neck [33,35], liver, lung, and abdominal cancers [29,61,127]. For example, the DL method has exhibited as an effective and clinically applicable tool for the segmentation of the head and neck anatomy for radiotherapy [34].…”

“…AI efforts in CT and MRI are already well underway and have demonstrated remarkable progress in various image analysis tasks [8][9][10][11]. In cancer screening, DL techniques have shown promise in CT screening for lung cancer and colonic polyps [28], MRI screening for prostate cancer [29], discriminating glioblastoma from brain metastasis with conventional MR images [30], breast cancer risk assessment with MR images [31,32], and segmentation of CT and MR images of head and neck (HN) cancer for MR-guided radiotherapy [33][34][35]. AI models trained on large datasets can extract high-dimensional representations, which show an increase in specificity compared with lower-dimensional machine learning methods often used in computer-aided detection software for lung cancer screening [36].…”

“…All major imaging vendors incorporate AI, which shows great promise for patient positioning [44], image acquisition, and reconstruction pipelines by reducing scan time, suppressing artifacts, and improving overall image quality via optimization of the signal-to-noise ratio (SNR) [45,46]. AI-based image reconstruction methods can also help minimize the radiation dose from CT images by improving image quality [23,35]. AI tools developed for specific, narrow tasks, such as case assignment, lesion detection, and segmentation of regions of interest, are critical for oncological imaging.…”

Essentially unedited deep‐learning‐based OARs are suitable for rigorous oropharyngeal and laryngeal cancer treatment planning

“…With the introduction of deep learning (DL) technology in radiotherapy, the performance of autosegmentation algorithms has improved substantially. [4][5][6][7][8][9] A substantial number of papers, of which we will only cite the reviews [10][11][12] and some recent efforts, [13][14][15][16][17] were devoted specifically to autosegmentation of HN organs at risk (OARs).As a logical first step, quality of OAR autosegmentation is often judged by geometric concordance metrics against the human-generated reference structures, [18][19][20][21][22] sometimes followed by a forward dosimetric analysis, for example, comparing the dose-volume histograms (DVHs) for the machine-generated structures against the manual ones, based on the original dose distribution. However, the real goal of autosegmentation is the inverse ability: the quality of a plan generated on the (unedited) autosegmented structures should be maintained if the manual OARs are substituted for the automatic ones during the analysis.…”

“…While it was suggested that accurate delineation of salivary and swallowing structures may not be necessary for satisfactory Head and Neck (HN) treatment planning, 3 the prevailing effort is clearly focused on improving the contours’ quality. With the introduction of deep learning (DL) technology in radiotherapy, the performance of autosegmentation algorithms has improved substantially 4–9 . A substantial number of papers, of which we will only cite the reviews 10–12 and some recent efforts, 13–17 were devoted specifically to autosegmentation of HN organs at risk (OARs).…”