MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Subset of Reproducible Radiomic Features as a Function of Multiple CT Imaging Parameters

M Shafiq ul Hassan1*, F Guo1 , H Chen3 , G Zhang4 , E Moros5 , Z Chen6 , (1) Yale New Haven Hospital, New Haven, CT, (2) Yale New Haven Hospital, New Haven, CT, (3) Yale New Haven Hospital, New Haven, CT, (4) H. Lee Moffitt Cancer Center, Tampa, FL, (5) H. Lee Moffitt Cancer Center, Tampa, FL, (6) Yale Univ. School of Medicine & YNHH, New Haven, CT

Presentations

(Sunday, 7/14/2019) 4:30 PM - 5:00 PM

Room: Exhibit Hall | Forum 9

Purpose: The extraction of quantitative information from images (Radiomics) has shown promising prospects for cancer prediction and diagnosis. The variation in CT imaging parameters directly affects the image quality and subsequently quantitative information, extracted from these images. In this work, the purpose is to identify a subset of reproducible features due to a large number of CT imaging parameters.

Methods: Total 56 scans of 3D printed texture inserts within Gammex phantom (RMI-465) were acquired on Siemens Force (28 scans) and GE Lightspeed (28 scans) CT scanners using a controlled scanning approach. The texture within the 3D printed inserts simulate the CT range and standard deviation typically found in human cancers. The scans were acquired using a number of image acquisition and reconstruction parameters such as kVp, mAs, pitch, FOV, reconstruction kernel and slice thickness. The sixty nine features including intensity (16), GLCM (26), GLRLM (11), GLSZM (11) and NGTDM (5) were extracted using an in-house program. The concordance correlation coefficient (CCC) is used to assess the reproducibility of features due to each individual parameter. Features having CCC > 0.90 and CCC < 0.90 are classified as reproducible and non-reproducible features.

Results: Second and high order features were sensitive to CT reconstruction parameters. Only 16 out of 53 texture features were reproducible with respect to reconstruction parameters (slice thickness, FOV and reconstruction kernel). Second order features describe relationship between neighboring voxels, therefore, greatly affected by the correlated noise. Higher order features (GLSZM, NGTDM) are sensitive to both acquisition and reconstruction kernels. Overall, 9 out of 69 features were found reproducible to studied CT imaging parameters.

Conclusion: Using a controlled scanning approach, a subset of radiomic features was identified. This subset of features could be used in subsequent clinical studies. In future, this particular approach will be expanded to 1100 features

Keywords

Data Acquisition, Quantitative Imaging, Reconstruction

Taxonomy

IM- CT: Phantoms - physical

Contact Email