3356 - Magnetic Resonance Imaging (MRI) Radiomics that Predicts the Treatment Outcome in a Phase 2 Trial of Ra-223 Combined with Stereotactic Body Radiation Therapy and Androgen Deprivation Therapy (ADT) for Oligometastatic Castrate Sensitive Prostate Cancer
Presenter(s)
Y. Huang1, T. Ketcherside2, A. Liu2, C. Han2, W. T. Watkins2, C. Hao3, H. Gilbertson2, S. V. Dandapani2, and K. Qing2; 1Jiangxi Cancer Hospital, Nanchang, China, 2Department of Radiation Oncology, City of Hope National Medical Center, Duarte, CA, 3Johns Hopkins University, Baltimore, MD
Purpose/Objective(s): In a single-center open-label phase II clinical trial investigating Ra-223 (Xofigo) combined with 36 weeks of ADT and stereotactic body radiation therapy (SBRT) for patients with oligometastatic castrate sensitive prostate cancer (CSPC, =5 metastases), the utility of pre-treatment MRI radiomics in prediction of post-treatment outcome was investigated.
Materials/Methods: A total of 13 patients with oligometastatic CSPC were included in this study. Each patient underwent T2-weighted (T2), T1-weighted (T1) and post contrast T1-weighted (T1C) MRI before and after (3-6 months) the treatment. A total number of 79 quantitative radiomic features including first order statistics, gray level cooccurrence matrix (GLCM), gray level run length matrix (GLRLM), grayl evel size zone matrix (GLSZM), and neighboring gray tone difference matrix (NGTDM) were processed using PyRadiomics (3D-Slicer). The utility of these radiomic features from pre-treatment and pre-post treatment changes to predict bone progression and systemic progression was investigated using lease absolute shrinkage and selection operator (LASSO) method.
Results: At data cutoff, all patients remained alive. Ten patients were found to have systemic progression, and 7 patients were found to have bone progression out of the 13 patients. From the pre-treatment radiomic features, only 1 feature (glrlm_LongRunLowGrayLevelEmphasis) from the T1C was found to be significantly different between groups with and without bone progression (P=0.025), and yielded an area under the curve (AUC) of 0.51. A total number of 58 features from T1 and 7 features from T2 were found to be significantly different between patients with systemic progression. Through LASSO analysis, 11 features from the pre-treatment T1 features were selected, and yielded an AUC of 0.88 to predict systemic progression. Features that provided large regression coefficients include glszm_SmallAreaEmphasis (0.26), glcm_JointEnergy (0.24), glcm_lmc2 (0.22) None of the T2 features were selected to yield sufficient predictive values by LASSO. From the pre/post changes of the radiomic features, 2 features from T1 (glcm_Correlation and first_order_Skewness) and 2 features from T2 (glcm_Correlation and glcm_Imc1) were found to be significantly different between patients with and without systemic progression. Through LASSO analysis, both T1 and T2 features were selected and yielded an AUC of 0.81 and 0.84 respectively to predict systemic progression. For bone progression, 4 features from T2 were initially significant, but LASSO selected only first_order_Skewness, yielding an AUC of 0.73.
Conclusion: In this study including 13 patients with oligometastatic CSPC underwent combined treatment of Xofigo, SBRT and ADT, pre-treatment T1 radiomics were found to have significant predictive values for future systemic progression, and these T1 radiomic features are mostly related to tumor heterogeneity and texture uniformity.