Project Intro
Vestibular schwannoma (VS) is a benign tumor with slow growth rate and clear boundary. Gamma knife (GK) surgery is a promising treatment for patients with VS. Taipei Veterans General Hospital (TVGH) is an internationally renowned treatment center for its reputation of GK. Thousands of GK cases have been collected in TVGH, including 500 VS cases of complete GK treatment. The center has gathered 250,000 MR images of before and after GKS of 6 times follow-ups. Most of the VS will shrink after two years of GK, but 52% of patients will encounter pseudo-progression, that VS is temporally swelling and afterword reducing, after GK surgery. Recently, the factors for predicting prognosis after the GK is still unclear. Furthermore, the high heterogeneity, and different tissue composition, growth rate, patterns in MR images, and response to GK of VS are observed in clinical practic. Therefore, exploring the effective predictors for pseudo-progression and efficient shrinkage after GK surgery is a key issue of clinical research in VS. This project is proposed to conduct in two years, and consist of three subprojects. The sub-project 1 is responsible for collecting and labelling clinical data of VS for sub-project 2 and 3, discussing prognostic factors after GK, and verifying the appropriateness of the prediction model developed from sub-project 2 and 3; sub-project 2 is responsible for automatically detecting VS’s lesion and surrounding tissue and building a prognostic prediction model for GK; sub-project 3 will analyze the radiomics of the image of VS, select the clinically significant radiomics features, and integrate the radiomics features into the prognosis prediction model for GK. The PI of main project and sub-project 2 (AI team) and the PI of sub-project 3 (radiomics team) are experienced scientists in MR principles, image processing and analysis of various brain diseases. They have cooperated closely with the team of sub-project 1 (clinical data team in TVGH) for more than 10 years and understand the clinical language that enables them to quickly catch the clinical insights and tackle the problem by appropriate approaches. The salient features of this project is the integration of selected dominate feature maps and image biomarkers resulted from convolutional neural network and radiomics, respectively, on MR images of multiple follow-ups. The aim is to construct two prediction models to assist clinical decisions: one is to predict the possible outcome before GK and suggest whether the GK should be conducted, another is to predict the change of VS volume (3 categories: volume changes 10% and volume reduces > 10% ) and side effect of hearing damage (3 to 5 categories from mild to severe) after the second year after GK. Subsequently, the models will assist the treatment planning of GK surgery and improve the development of precision medicine.