MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Modeling and Recovering Gray-Level Co-Occurrence-Based Radiomics in the Presence of Blur and Noise

G Gang1*, J Stayman2, (1) Johns Hopkins University, Baltimore, MD, (2) Johns Hopkins University, Baltimore, MD

Presentations

(Sunday, 7/12/2020) 3:30 PM - 4:30 PM [Eastern Time (GMT-4)]

Room: Track 1

Purpose: While there is increasing interest in the application of data science and radiomics to large patient databases, inherent variability in these databases confounds potential scientific inferences. This is particularly true in x-ray computed tomography where image properties (and hence radiomics) depend highly on x-ray technique, scan protocol, patient size and habitus, device vendor, etc. In this work, we identify one popular class of radiomics based on gray-level co-occurrence matric (GLCM) and model the effect of blur and noise on GLCM quantitation. With this new model, we attempt to recover the original target GLCM based on noisy, blurry measurements.

Methods: Our mathematical model of GLCM presumes a known shift-invariant blur and known noise distribution of measurement data (e.g. through direct measurement or system modeling). Such a model applies to imaging scenarios that are locally stationary/shift-invariant. While the dependency of GLCM on blur is image-dependent and requires explicit deconvolution, one can show that noise broadens the GLCM – specifically, convolving the GLCM of the noise alone with the noiseless image GLCM. These observations allow both modeling and potential recovery of the original GLCM, which we illustrate in simulation studies.

Results: We find that it is possible to recover GLCM estimates of the original unblurred and no-noise images. Specifically, we find good agreement between recovered GLCM values and the target GLCM; however, GLCM estimates themselves are subject to increased noise due to the need for deconvolution.

Conclusions: A mathematical model of how blur and noise affects GLCM provides not only improved understanding of the underlying variability in GLCM-based radiomics but also provides a potential avenue for reducing this variability in large databases that have varying noise and resolution properties. While this work concentrates on GLCM metrics, similar analysis may be possible for other radiomic measures providing additional opportunities for standardization.

Funding Support, Disclosures, and Conflict of Interest: Funding from NIH, Siemens Healthineers, Fischer Imaging, Canon Medical Research Partnerships with GE Healthcare, Philips, United Imaging

Keywords

Not Applicable / None Entered.

Taxonomy

IM- CT: Radiomics

Contact Email