Program Information
The Predictive T1 MR Radiomics Features for the Outcome of Locally Advanced Rectal Cancer Patients
H Zhong1*, P Boimel2 , Y Fan3 , J Wang4 , H Geng5 , M Huang6 , C Cheng7 , M Rosen8 , E Ben-Josef9 , Y Xiao10 , (1) University of Pennsylvania, Philadelphia, PA, (2) University of Pennsylvania, Philadelphia, PA, (3) University of Pennsylvania, Philadelphia, PA, (4) Fudan University Shanghai Cancer Center, Shanghai, ,(5) University of Pennsylvania, Bryn Mawr, Pennsylvania, (6) University of Pennsylvania, Philadelphia, PA, (7) University of Pennsylvania, Philadelphia, PA, (8) University of Pennsylvania, Philadelphia, PA, (9) University of Pennsylvania, Philadelphia, Pennsylvania, (10) University of Pennsylvania, Philadelphia, PA
Presentations
SU-I-GPD-J-96 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose: To study the possible correlation between T1 Magnetic Resonance (MR) radiomics features and rectal cancer prognosis, and to compare the performance of unenhanced radiomics features, enhanced radiomics features, and their differences.
Methods: A total of 30 locally advanced rectal cancer patients treated with neoadjuvant chemoradiotherapy (CRT) were enrolled in this study. All patients underwent T1 unenhanced and enhanced MR scans before CRT. The primary lesions of these patients were contoured on T1 enhanced MR images by the expert radiation oncologists. The gray levels of all MR images have been normalized by standard score (z-score). A total of 170 radiomics features were extracted from the filtered images, including three geometry features, 12 histogram features, 100 GLCM texture features, and 55 GLRLM texture features. The difference between enhanced features and unenhanced features has been calculated. The T-tests were used for feature selection. The top 3 selected features were used in Cox Regression model for predicting the metastasis-free survival and progression-free survival of the patients.
Results: 10 of the 30 patients develop metastasis, 11 of them have progression. The t-test shows unenhanced T1 features, enhanced T1 features, and their differences do not correlate with metastasis (threshold: p<0.05). For progression-free survival, no correlating feature was found for unenhanced CT, while three features for enhanced CT and 36 features of their differences are found to correlate significantly. The Cox regression model for predicting metastasis-free survival achieved testing C-indices of 0.51, 0.64, and 0.65 for unenhanced features, enhanced features, and their differences respectively. For progression-free survival, the C-indices are 0.56, 0.66, 0.67 for unenhanced features, enhanced features, and their differences respectively.
Conclusion: We extracted predictive radiomics features from enhanced T1 MR for the outcome of rectal cancer patients. The features from the difference are more predictive than features from enhanced images alone.
Contact Email: