Program Information
The Effect of Reducing Milliamp Seconds On Computed Tomography Radiomics Features
D Mackin1*, R Ger1 , X Fave1 , L Zhang1 , J Yang1 , S Bache2 , P Chi1 , A Jones1, C Dodge3,, L Court1 (1) The University of Texas MD Anderson Cancer Center, Houston, TX, (4) Texas Children's Hospital, Houston, TX
Presentations
TU-H-FS4-3 (Tuesday, August 1, 2017) 4:30 PM - 6:00 PM Room: Four Seasons 4
Purpose: Computed tomography radiomics feature values are sensitive to scan acquisition parameters. Of particular concern is milliamp-seconds (mAs) because this value cannot be changed for retrospective reconstruction and because the range of values commonly used is wide. We sought to determine which features are sensitive to changes in mAs and the change required to affect feature values.
Methods: A radiomics phantom was imaged 12 times with mAs values ranging from 25 to 300 and all other parameters held constant. 41 features were calculated from the categories of intensity direct (n=11), neighborhood intensity difference (n=5), gray-level co-occurrence matrix (n=21), and gray-level run length (n=4). Each feature was calculated from four materials (wood, acrylic, rubber, and cork) with four different image preprocessing methods: no preprocessing, bit-depth resampling, Butterworth smoothing, and Butterworth smoothing + bit-depth resampling. To gauge the effects of mAs variation, the change of each of the features was scaled by the corresponding coefficient of variation from tumor samples from patients with non-small cell lung cancer.
Results: Changes in feature values were larger in materials with less texture. Reducing the mAs from 300 to 150 significantly changed the values of 33 of 41 features calculated for acrylic and 3 of 41 features calculated for wood. However, none of the features demonstrated a clear mAs dependence for rubber or cork, materials with texture similar to the textures found in human tissues. Image smoothing and bit-depth resampling reduced the mAs dependence for 16 of 20 features with mAs dependence in wood. Preprocessing had inconsistent effects for the other materials.
Conclusion: The effects of changing mAs depend on the texture. For materials with little texture, small reductions produced sizable changes in feature values. For materials with tissue-like texture, even large changes in mAs produced negligible changes in the feature values.
Contact Email: