Volumetric Assessment of Vestibular Schwannomas Exhibiting Internal Growth After Stereotactic Radiosurgery: Application of Artificial Intelligence.
Authors
Affiliations (9)
Affiliations (9)
- Yale School of Medicine, Yale University, New Haven, United States.
- Yale School of Medicine, Department of Orthopedics and Rehabilitation, Yale University, New Haven, United States.
- School of Medicine, University of Sheffield, Sheffield, United Kingdom.
- Yale School of Medicine, Department of Surgery, Division of Otolaryngology-Head and Neck Surgery, Yale University, New Haven, United States.
- Yale School of Medicine, Department of Therapeutic Radiology, Yale University, New Haven, United States.
- Yale School of Medicine, Department of Neurosurgery, Yale University, New Haven, United States.
- Yale School of Medicine, Department of Psychiatry, Yale University, New Haven, United States.
- Yale School of Medicine, Department of Radiology and Biomedical Imaging, Yale University, New Haven, United States.
- Yale School of Medicine, Department of Surgery, Division of Otolaryngology-Head and Neck Surgery, Yale University, New Haven, United States. [email protected].
Abstract
Vestibular schwannomas (VS) are monitored for growth after stereotactic radiosurgery (SRS) using serial MRI. Conventional measurements often miss internal regrowth within necrotic/cystic tumor components-a subtle and difficult-to-detect imaging scenario that may indicate treatment failure. We evaluated a custom AI-based segmentation tool to detect internal tumor regrowth after SRS, comparing it with expert segmentation for spatial agreement, volumetric change, and radiographic failure classification (≥ 20% volume increase). We retrospectively analyzed 170 MRI scans from 85 patients with unilateral VS treated with SRS at a tertiary academic center (2001-2023). Tumors were segmented using expert manual methods and a lab-developed convolutional neural network (CNN)-based AI model. Tumors with internal regrowth comprised 4.7% of cases yet accounted for 27% of radiographic failures. Within the internal regrowth subset, the AI model achieved a median Dice similarity coefficient of 0.85, demonstrating favorable agreement with expert segmentation. Concordance between AI- and manually derived volumetric change was favorable (concordance correlation coefficient, 0.87), with a median percent difference of 9.29% (range, 3.57-20.78%). AI classification of radiographic failure within the internal regrowth subset was consistent with expert assessment, and performance remained comparable to the broader cohort. Internal tumor regrowth after SRS is a rare and potentially underrecognized post-treatment imaging pattern associated with radiographic failure that may be difficult to identify using conventional radiographic assessment. In this preliminary study, agreement between AI and expert assessment highlights the potential role of AI-based volumetric tools in improving recognition and longitudinal surveillance of structurally complex post-treatment tumor changes.