MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Impact of Quantization Parameters On Radiomics Feature Variation in Low Field Strength Magnetic Resonance Images

G Simpson1*, J Ford2, F Yang2, N Dogan2, (1) University of Miami, Coral Gables, FL,(2) University of Miami Miller School of Medicine, Miami, FL

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: Low field strength magnetic resonance (MR) set-up images are hypothesized to contain patient specific disease information extractable using radiomics feature analysis. The potential prognostic power of radiomics feature analysis using longitudinal MRI data is exciting. However, the low signal-to-noise ratio of these images calls into question the repeatability of features. The purpose of this study is to investigate the variation of radiomics features extracted from a texture phantom through variation of quantization algorithm and number of bin levels.

Methods: Images were acquired on a 0.35T hybrid MR/RT of a daily image QA phantom with inserted texture test tubes. Texture was produced by filling each tube with vitamin E-pills, rolled gauze, capillary tubes, or cut-up plastic intravenous tubing, then brimming with copper sulfate-doped water. Radiomics features belonging to gray-level co-occurrence matrix (GLCM), gray-level run length matrix (GLRLM), gray-level size zone matrix (GLSZM), or neighborhood gray-tone difference matrix (NGTDM) were extracted from 37 daily images. Radiomics features were calculated using combinations of three quantization algorithms (Lloyd-Max, histogram equalization, or uniform probability) and three bin levels (64, 128, or 256). The coefficient of variation (CV%) was then calculated.

Results: The number of radiomics features with median CV% across all four texture inserts below 10% for Lloyd-Max with 64 bins is 11 features, 12 features with 128 levels, and 13 features for 256 levels. Use of a histogram equalization method with 64 levels yielded 26 features below 10%, 18 features with 128 levels, and 256 levels resulted in 12 features. Use of a uniform quantization with 64 or 128 levels resulted in 12 features below 10%, and 13 features when using 256 levels.

Conclusion: This study demonstrated the most repeatable combination for radiomics features is histogram equalization method with 64 levels. However, the repeatability and prognostic ability need evaluation in a clinical setting.

Keywords

Texture Analysis, Image Analysis, Phantoms

Taxonomy

Not Applicable / None Entered.

Contact Email