MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Radiomics Feature Robustness as Measured From An MRI Radiomics Phantom

J Lee*, A Steinmann , y ding , H Lee , J Wang , D Followill , L Court , MD Anderson Cancer Center, Houston, TX

Presentations

(Monday, 7/30/2018) 1:15 PM - 1:45 PM

Room: Exhibit Hall | Forum 8

Purpose: To characterize a new MRI radiomics phantom by investigating feature variability across various protocols and scanners.

Methods: An MRI phantom consisting of 20 different MR visible materials was constructed. T1- and T2-weighted images of the phantom were acquired on a Philips and Siemens 1.5T MR scanner while varying the acquisition parameters. We extracted 90 radiomics features from the phantom images using IBEX software. First, the suitability of each phantom material was evaluated by comparing each feature value within the two-sigma range of feature values extracted from 97 T1- and T2-weighted MR images of patients with brain lesions. Then the inter- and intra-scanner variability in phantom features were evaluated for both T1- and T2-weighted images. The phantom material feature variability (defined as the range divided by the mean) was considered to be acceptable if it was less than half of the coefficient of variation of that feature from the patient cohort.

Results: An average of 92% and 59% of the phantom radiomics features were within the established patient population bounds for T1 and T2 weighted images, respectively. When considering intra-scanner variability, we found that an average of 68.5% and 71.7% of the phantom radiomics features varied less than inter-patient variation for T1 and T2 scans, respectively. For inter-scanner variability, an average of 33.7% and 76.7% of the phantom radiomics features varied less than inter-patient variation for T1 and T2 scans, respectively.

Conclusion: The majority of radiomics features extracted from images of our MRI radiomics phantom were within the range of patient feature values. We then used this phantom to identify which features were robust to changes in scanning protocol and scanner manufacturer when compared to inter-patient variations.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email