0013

View Abstract / View Presentation

Using MR Radiomics to Improve Prediction of Local Tumor Control after Radiosurgery in Brain Metastases

Chien-Yi Liao¹, Cheng-Chia Lee^2,3,4, Huai-Che Yang^2,3, Wen-Yuh Chung^2,3, Hsiu-Mei Wu^3,5, Wan-Yuo Guo^3,5, Ren-Shyan Liu^1,6,7, and Chia-Feng Lu^1,8
¹Department of Biomedical Imaging and Radiological Sciences, National Yang-Ming University, Taipei, Taiwan, Taipei, Taiwan, ²Department of Neurosurgery, Neurological Institute, Taipei Veteran General Hospital, Taipei, Taiwan, Taipei, Taiwan, ³School of Medicine, National Yang-Ming University, Taipei, Taiwan, Taipei, Taiwan, ⁴Brain Research Center, National Yang-Ming University, Taipei, Taiwan, Taipei, Taiwan, ⁵Department of Radiology, Taipei Veteran General Hospital, Taipei, Taiwan, Taipei, Taiwan, ⁶Department of Medical Imaging, Cheng-Hsin General Hospital, Taipei, Taiwan, Taipei, Taiwan, ⁷Molecular and Genetic Imaging Core, Taiwan Animal Consortium, Taipei, Taiwan, Taipei, Taiwan, ⁸Institute of Biophotonics, National Yang-Ming University, Taipei, Taiwan, Taipei, Taiwan

The prediction of treatment response after Gamma Knife stereotactic radiosurgery (GKRS) can benefit patient management. We suggested that imaging characteristics extracted from preradiosurgical MRIs combined with clinical information can effectively predict local tumor control.

Figure 1. Image processing steps and radiomics flowchart

(a) Image acquisition of T1cwc, T1w and T2w images, registration of T1w and T2w to T1c images and intensity normalization. (b) ROI delineation for the treatment planning of GKRS by experienced neurosurgeons and radiologists on all MRI slices covering tumor regions. (c) Extraction of radiomic features from the tumor ROIs on the MRIs.

Figure 2. Representative cases and the model performance in predicting local tumor control

(a) T1c images and reconstructed 3D tumor models. (b) The SVM scores in two representative cases based on the combination of radiomic and clinical features. (c) Receiver operating characteristic curves and the area under the curves (AUCs) of three SVM classification models.