Program Information
The Study of Temporal Consistence for Radiomics Features
P Hu, J Wang , y sun , W Hu , Z Zhang* , Fudan University Shanghai Cancer Center, Shanghai, shanghai
Presentations
MO-RPM-GePD-IT-4 (Monday, July 31, 2017) 3:45 PM - 4:15 PM Room: Imaging ePoster Theater
Purpose: To evaluate the temporal consistence of radiomics features by using rabbits with VX2 tumor tissue. Choose stable radiomics features to establish predictive and prognosis model.
Methods: The VX2 tumor tissue was orthotopically transplanted into the rectal wall of the New Zealand white rabbits. Four New Zealand white rabbits were transplanted. Since four weeks after tumor implantation, we scanned one rabbit by MR a week. After MR scanning, the rabbit was killed. Totally, four MR scans were obtained for four rabbits, respectively. The rectal GTV was distinguished and segmented by an experienced radiation oncologist in MIM software, the delineation was double-checked and the non-invaded rectal wall and the air inside the rectum were carefully excluded. 252 radiomics features were defined in our study by using QIAT, an in-house developed radiomics feature extraction software, to calculate radiomics features for analysis. The relationship between MR radiomics features and time was evaluation by calculating linear correlation factor.
Results: The volumes of rabbits at each week follow the exponential relationship, and the value of R2 is 0.9964. 94 out of 252 features showed high leaner correlation (R2≥0.9) with time. 107 out of 252 features showed medium leaner correlation (0.5≤R2＜0.9), 51 out of 252 features showed low leaner correlation (R2＜0.5). 79 out of the 94 features showed large deviation, 15 out of the 94 features showed low deviation.
Conclusion: 37.3% radiomics features have a large temporal inconsistence. Since radiomics features are highly time dependent, it should careful when using radiomics features to establish predictive and prognosis model. Further analysis of radiomics can be warranted for treatment monitoring and prognosis prediction by using these 15 radiomics features which showed low deviation.
Contact Email: