Segmentation of Vestibular Schwannomas on Postoperative Gadolinium-Enhanced T1-Weighted and Noncontrast T2-Weighted Magnetic Resonance Imaging Using Deep Learning

Abstract

Surveillance of postoperative vestibular schwannomas currently relies on manual segmentation and measurement of the tumor by content experts, which is both labor intensive and time consuming. We aimed to develop and validate deep learning models for automatic segmentation of postoperative vestibular schwannomas on gadolinium-enhanced T1-weighted magnetic resonance imaging (GdT1WI) and noncontrast high-resolution T2-weighted magnetic resonance imaging (HRT2WI). Study Design: A supervised machine learning approach using a U-Net model was applied to segment magnetic resonance imaging images into pixels representing vestibular schwannoma and background pixels. Setting: Tertiary care hospital. Patients: Our retrospective data set consisted of 122 GdT1WI and 122 HRT2WI studies in 82 postoperative adult patients with a vestibular schwannoma treated with subtotal surgical resection between September 1, 2007, and April 17, 2018. Forty-nine percent of our cohort was female, the mean age at the time of surgery was 49.8 years, and the median time from surgery to follow-up scan was 2.26 years. Intervention(s): N/A.Main Outcome Measure(s): Tumor areas were manually segmented in axial images and used as ground truth for training and evaluation of the model. We measured the Dice score of the predicted segmentation results in comparison to manual segmentations from experts to assess the model's accuracy.Results: The GdT1WI model achieved a Dice score of 0.89, and the HRT2WI model achieved a Dice score of 0.85. Conclusion:We demonstrated that postoperative vestibular schwannomas can be accurately segmented on GdT1WI and HRT2WI without human intervention using deep learning. This artificial intelligence technology has the potential to improve the postoperative surveillance and management of patients with vestibular schwannomas.

Current State of Artificial Intelligence in Clinical Applications for Head and Neck MR Imaging

“…73 In contrast, Korte et al investigated the delineations of the parotid gland, submandibular gland, and neck lymph nodes by using a T2WI dataset with multiple 3D-U-Net systems, resulting in DSC values of approximately 0.8. 74 The segmentation of other lesions and anatomical structures such as vestibular schwannoma in the cerebellopontine angle, 75,76 the inner ear and its related structures (e.g., cochlea, vestibule), [77][78][79] and venous malformations of the neck 80 has been described. Segmentation of the inner ear and its related structures necessary for the diagnosis of endolymphatic hydrops has also been a concern (see the section below titled 'Disease classification and diagnosis') and would be valuable for clinical practice.…”

Deep Learning Techniques and Imaging in Otorhinolaryngology—A State-of-the-Art Review

“…AI-assisted radiomics can be extremely useful in the follow-up of specific diseases, such as vestibular schwannoma, whose surveillance is nowadays performed through analogical segmentation and an analysis of serial MRI scans to detect tumor enlargement. A deep learning approach can be applied for tumor detection and segmentation in treatment-naïve patients [120], both after radiosurgery [121], in evaluating residual disease [122], and in predicting tumor enlargement based on radiomics parameters during follow-up [123].…”