Prediction of transient tumor enlargement using MRI tumor texture after radiosurgery on vestibular schwannoma

TitlePrediction of transient tumor enlargement using MRI tumor texture after radiosurgery on vestibular schwannoma
Publication TypeJournal Article
Year of Publication2020
AuthorsLangenhuizen, PPJH, Sebregts, SHP, Zinger, S, Leenstra, S, Verheul, JB, With, PHN
JournalMedical Physics
Volume471195
Issue4
Pagination1692 - 1701
Date PublishedJan-04-2020
Type of ArticleJournal Article
ISSN0094-2405
Abstract

Introduction
Gamma Knife radiosurgery (GKRS) is a well-established treatment for small- to medium-sized vestibular schwannomas (VS). However, this treatment is controversial for larger VS. One of its drawbacks is that VS can present a radiation-induced transient tumor enlargement (TTE). For larger VS, such a swelling may cause symptoms related to mass effect, necessitating microsurgery. Currently, it is not possible to predict this adverse effect. We evaluated the predictability of TTE by quantitatively analyzing the tumor appearance on MRI. The goal is to determine the optimal treatment strategy, i.e. radiosurgery or microsurgery, on an individual basis.

Methods
From our database, patients with large VS (>4cc) and minimum follow-up of three years, were identified. The TTE classification was based on evaluation of MRI scans at 6, 12, 24 and 36 months, according to strict volumetric criteria. We evaluated the influence of MRI tumor texture characteristics on TTE. These texture characteristics were quantified by calculating features based on gray-level co-occurrence matrices (GLCM), computed on T1-weighted, T2-weighted, and T1-weighted contrast-enhanced MRIs. Correlation was determined between these characteristics and TTE using machine-learning methods.

Results
Between 2002 and 2015, 795 VS patients received GKRS as primary treatment at our center. The strict criteria for TTE and non-TTE led to the inclusion of 67 patients, of which 26 exhibited TTE. By employing GLCM-based features, we developed a model to predict TTE. We obtained a prediction sensitivity and specificity of 83% and 79%, respectively, using Support Vector Machines. These results improved for larger tumor volumes, i.e. in 7cc or larger, the results obtained were 85% and 87%, respectively.

Conclusion
Results from this research clearly show that MRI differences in VS tumor texture can be exploited to predict TTE in large VS. The developed prediction model can lead to an optimal treatment strategy selection on an individual basis.

URLhttps://onlinelibrary.wiley.com/doi/abs/10.1002/mp.14042
DOI10.1002/mp.v47.410.1002/mp.14042
Short TitleMed. Phys.