Cardiac SPECT radiomic features repeatability and reproducibility: A multi-scanner phantom study

Abstract

Background:The aim of this study was to assess the robustness of cardiac SPECT radiomics features against changes in imaging settings including acquisition and reconstruction settings. Methods: Four scanners were used to acquire SPECT scans of a cardiac phantom with 5mCi of 99m Tc. The effects of different image acquisition and reconstruction settings including the Number of View, View Matrix Size, attenuation correction, image reconstruction algorithm, number of iterations, number of subsets, type of filter, full width at half maximum (FWHM) of Gaussian filter, Butterworth filter order, and Butterworth filter cut-off were studied. In total 5263 different images were reconstructed. Eighty-seven radiomic features including first, second, and high order textures were extracted from images. To assess reproducibility and repeatability the coefficient of variation (COV) was used for each image feature over the different imaging settings. Result: IDMN and IDN features from GLCM, RP from GLRLM, ZE from GLSZM, and DE from GLDM feature sets were the only features that were the most reproducible (COV ≤ 5%) against changes in all imaging settings. In addition, the IDMN feature from GLCM, LALGLE, SALGLE and LGLZE from GLSZM, and SDLGLE from GLDM feature sets were the features that were less reproducible (COV>20 %) against changes in all imaging settings. Matrix size has the greatest impact on feature variability as most of features are not repeatable and 82.76% of them had (COV>20 %). Conclusion: Repeatability and reproducibility of SPECT/CT radiomics texture features in different imaging settings is feature-dependent, and different image acquisitions and reconstructions have different effects on radiomics texture features. Low COV radiomics features could be consider for further clinical studies.

Repeatability of radiomic features in magnetic resonance imaging of glioblastoma: Test–retest and image registration analyses

“…Although radiomic analyses are becoming increasing mature, there are a number of important technical limitations, and many radiomic features are vulnerable to significant variations based on image acquisition, reconstruction, and processing methods, as reported by ongoing radiomics studies. [23][24][25][26][27] Moreover, as hundreds of feature sets are available for consideration in medical imaging, it is necessary to consider the reproducibility and repeatability of radiomic features as a feasible measure to preselect features for further analysis, such as classification and clinical correlation. 23 In image biomarkers development, there are two main frontiers which should be assessed in regard to robustness of radiomic features.…”

“…Variations in these main steps and their substeps may result in notable alterations in radiomic features as considered for final outcome analysis. Although radiomic analyses are becoming increasing mature, there are a number of important technical limitations, and many radiomic features are vulnerable to significant variations based on image acquisition, reconstruction, and processing methods, as reported by ongoing radiomics studies 23–27 . Moreover, as hundreds of feature sets are available for consideration in medical imaging, it is necessary to consider the reproducibility and repeatability of radiomic features as a feasible measure to preselect features for further analysis, such as classification and clinical correlation 23 …”

Exploration of an Integrative Prognostic Model of Radiogenomics Features With Underlying Gene Expression Patterns in Clear Cell Renal Cell Carcinoma

“…A gray level zone is described as the number of connected voxels which show the same intensity. The texture feature Large Area High Gray Level Emphasis from GLSZM quantifies the proportion in the image of the joint distribution of smaller size zones with higher gray-level values, which has been formerly adopted in the assessment of the robustness or patient response in different imageological examinations ( 26 , 27 ). The GLDM-based textural feature Gray Level Non Uniformity (GLN) calculates the similarity of gray-level intensity values, where a lower GLN refers to a higher intensity value in the image ( 28 ).…”

Left ventricular mechanical dyssynchrony in patient with CAD: The Saga continues

“…Therefore, I foresee a risk calculator that takes into account all these variables to predict the likelihood of CAD, significant ischemia, and provide prognostic value. With the advancement of radiomics and artificial intelligence networking (currently being evaluated in SPECT imaging), 24,25 I believe we should start taking advantage of the technology, find algorithms, combine these parameters and many more patterns identified by artificial intelligence, to identify patients with high-risk MPI and guide clinical decisions. At the end of the day, the stress MPI is ordered for a clinical reason and therefore, we should take advantage of all the parameters it offers to help the clinician make an informed decision.…”