Encrypted login | home

Program Information

Radiomics Feature Variability On 0.35T MR-Guided-RT System

no image available
K Padgett

K Padgett, I Mihaylov, University of Miami: Miami, FL


TU-L-GePD-JT-1 (Tuesday, August 1, 2017) 1:15 PM - 1:45 PM Room: Joint Imaging-Therapy ePoster Theater

Purpose: To explore the variability of the texture features derived from gray-level co-occurrence matrices (GLCMs) on a 0.35T MR scanner utilized for MRIGRT.

Methods: An ACR-MR accreditation phantom was scanned six times over a period of three months on the ViewRay 0.35T MRI system. The reconstructed datasets had a resolution of 1.5x1.48x1.48 mm3. Six regions of interest (ROIs) were defined on one scan and propagated to the other five scans through registration. Two ROIs were placed in areas with uniform intensity levels, centrally and one peripherally placed. Two other ROIs were centered in the geometric distortion module such that one covered only the central region, while the other larger ROI included the center and periphery, thereby probing the effects of image non-uniformity. The last two ROIs included phantom sections which exhibited fairly distinct texture. For each ROI 256-bin GLCM was created for each of the six scans. The GLCMs covered the entire range of intensities specific for each particular ROI. Eighteen commonly used image features were calculated form the GLCMs and the variability (defined as standard deviation divided by the average) of those features was assessed.

Results: The largest variation was observed for the two uniform regions, where majority of the features varied within 40%. Majority of the features for the ROIs with distinct texture varied within 20% to 30% range. The central region ROIs (also with distinct texture pattern) varied at 20% to 25% level, where the larger ROI including the periphery of the phantom exhibited larger variations.

Conclusion: The uniform regions ROIs yield textural features which show larger variation. The overall performance of the 0.35T MRI system is of the order of 20% to 30%. These numbers are probably lower limits on the radiomics features variability expected for realistic clinical situations.

Contact Email: