MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Stability and Reproducibility Study of Radiomics Features Using a 3D-Printed Biological Phantom

K Nie, N Yue, X Wang, Y Li*, Rutgers Cancer Institute of New Jersey, New Brunswick, NJ

Presentations

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

Room: AAPM ePoster Library

Purpose: To study the stability and reproducibility of CT radiomics features using a novel 3D-printed biological phantom.


Methods: A radiomics phantom consisting of various materials (plaster bricks, cork, wood, polystyrene, solid water and water), 3D-printed texture patterns and biological tissue and bone, was constructed. The phantom was scanned using varied tube currents and voltages, orientation, slice thickness, image resolution, and acquisition modes within 15 minutes and then scanned at multiple time points with one protocol on a single CT. A total of 19 shape-based, 105 histogram-based and 102 texture-based radiomics features were extracted using IBEX. The reproducibility and stability for all the features were evaluated based on the relative standard variation as well as the concordance correlation coefficient (CCC). We define 0.1% and 30% as the lower and upper bounds of stability and reproducibility for the relative standard deviation and 0.9 and 0.1 for the CCC.


Results: Overall, good reproducibility was observed in all components of our phantom with some material-dependent variations: the best observed (222 out of 226 features) in tissue and bone while the worst in polystyrene and fine plaster (154/226). All features showed great stability with different CT acquisition modes, tube current or voltage changes but compromised stability with varied scan orientation, image resolution and slice thickness. Second-order GLCM-based texture features are more sensitive to image protocols variations compared to the gradient-based or the first-order histogram-based features.


Conclusion: Standardization of the image protocol is absolutely necessary especially for analysis relying on higher order features and a stricter constrain should be placed on those image parameters affecting resolution. Being highly reproducible in the intra-scanner test-retest, our biological phantom is an ideal platform for benchmarking the inter-scanner variation.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email