MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Repeatability of Radiomic Texture Features Using Low Field Strength MR Images

G Simpson1*, J Ford1 , N Dogan2 , (1) University of Miami, Miami, FL, (2) University of Miami Miller School of Medicine, Miami, FL

Presentations

(Wednesday, 7/17/2019) 10:30 AM - 11:00 AM

Room: Exhibit Hall | Forum 8

Purpose: MR-guided radiotherapy devices have the ability to accurately track patient position and produce images with superior soft tissue contrast to conventional onboard x-ray imaging. Daily setup images have stirred excitement for the prospect of utilizing radiomic texture analysis for response assessment. The objective of this study is to evaluate the repeatability of MRI-based texture features extracted from a phantom and spinal cord of five liver stereotactic body radiation therapy (SBRT) patients.

Methods: MR images were acquired with a 0.35T hybrid MR/RT device. 39 features belonging to gray-level co-occurrence matrix (GLCOM), gray-level run length matrix (GLRLM), gray-level size zone matrix (GLSZM), or neighborhood gray-level difference matrix (NGTDM) were extracted from 37 phantom images. Texture was created in the phantom by filling test tubes with either vitamin E pills, gauze, capillary tubes, or with cut plastic intravenous tubing, then brimmed with water. Textures were then extracted from spinal cord sections of each patient (4 or 5 fractions). The coefficient of variation (CV) of each feature was calculated (JMP Pro, SAS Institute, Cary, NC) for each volume of interest.

Results: The CV of GLCOM-based features Entropy, Homogeneity, and Sum Average was below 10% across all phantom structures. Within GLSZM, Small Zone Emphasis (SZE), Zone Percentage (ZP), and Zone Size Non-uniformity (ZSN) also achieve a CV < 10%. GLRLM Long Run Emphasis (LRE), Run-Length Non-uniformity (RLN), Run Percentage (RP), and Short Run Emphasis (SRE) had a CV < 10%. Features extracted from volumes of spinal cord which also achieved a CV < 10%, averaged over the 5 patients, were Entropy, SZE, ZP, LRE, RP, RLN, and SRE.

Conclusion: These preliminary results suggest some features are repeatable using the clinical imaging protocol in the phantom and in vivo. These results warrant further study of radiomic texture analysis in low-field MR images.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email