Program Information
Reproducibility with Repeat MR in Radiomics Study for Rectal Cancer
P Hu, J Wang , y sun , J Gan , R Luo , W Hu , Z Zhang* , Fudan University Shanghai Cancer Center, Shanghai, shanghai
Presentations
MO-RPM-GePD-IT-2 (Monday, July 31, 2017) 3:45 PM - 4:15 PM Room: Imaging ePoster Theater
Purpose: To evaluate the reproducibility of radiomics features by test-retest and contour-recontour magnetic resonance ₍MR₎ scans in rectal cancer. To choose stable radiomics features for rectal cancer.
Methods: Test-retest MR images of 11 rectal cancer patients were enrolled in this study, each of whom underwent two MR scans within average 18.6 days. 256 radiomics features were defined in our study by using QIAT ₍Quantitative Image Analysis Toolbox₎, an in-house developed radiomics feature extraction software, to calculate radiomics features for analysis. These features were divided into eight groups: 1 gray level co-occurrence matrix ₍GLCM₎, 2 gray level run-length matrix ₍GLRLM₎, 3 Histogram, 4wavelet GLCM, 5 wavelet GLRLM, 6 wavelet histogram, 7 sharp and 8 fractal dimension. The concordance correlation coefficient ₍CCC₎ of two MR scans was calculated to assess the reproducibility. Contour-recontour MR images of 20 rectal cancer patients previously enrolled in a clinical trial were retrospectively obtained and used to determine CCC for 256 different radiomics features.
Results: The result of contour-recontour, 92 out of 256 features showed high reproducibility (CCC≥0.9), 113 out of 256 features showed medium reproducibility (0.9＞CCC≥0.5) and 51 out of 256 features showed low reproducibility (CCC＜0.5). The result of test-retest, 2 out of 256 features showed high reproducibility (CCC≥0.9), 79 out of 256 features showed medium reproducibility (0.9＞CCC≥0.5) and 175 out of 256 features showed low reproducibility (CCC＜0.5).
Conclusion: Most radiomics features are robust to the different radiation oncologists’ contouring. As a longer wait time, most of radiomics features showed low reproducibility. It needs more analysis and more careful to establish predictive model when using different patients with different stages.
Contact Email: