Computed tomography radiomic texture features dependence upon imaging parameters
Loading...
Date
2019-11
Authors
Makosa, Frank
Journal Title
Journal ISSN
Volume Title
Publisher
University of the Free State
Abstract
Introduction and Aim: Few studies have been carried out to determine the influence of Computed Tomography (CT) acquisition parameters (slice thickness, tube potential difference (kVp), and tube current time product (mAs)) on the quantitative image features in radiomics studies. There is little evidence in the published literature, of studies that use mathematics to establish radiomic texture features that are independent of the CT scan technique parameters. The stability of radiomic texture features may have a great impact on the diagnosis and treatment of cancers. Robust texture features can be used to track radiotherapy treatment response. In this study radiomic texture features were investigated to identify features that did not depend on the CT technique parameters. Methodology: The credence cartridge radiomics (CCR) phantom was imaged at four CT units at the Universitas Academic and the National District hospitals. The tube current-time product (mAs) was varied from 75 to 400 mAs in steps of 25mAs while the kilovoltage peak and slice thickness were kept set at 120kVp and 5 mm respectively. The CT tube potential was investigated at 80, 100, 120 and 135 kVp whilst mAs and slice thickness was kept set at 300 mAs and 5 mm respectively. The slice thickness was varied from 1 mm to 5 mm whilst the mAs and kVp was kept constant at 300 mAs and 120kVp respectively. The acquisition field of view (FOV) and pitch were kept constant. The images obtained were processed using PyRadiomics software platform of 3D Slicer and the Matlab 2017a package. PyRadiomics was used to segment and extract a total of 105 radiomics texture features for each region of interest (ROI) delineated on an image. The 105 radiomic features included 13 shape features, 18 first order statistics features, 23 grey-level co-occurrence matrix, 14 grey level difference matrix, 16 grey-level run length matrix, 16 grey level size zone matrix and 5 neighbourhood grey tone difference matrix features. For each 10 CCR phantom inserts, 16 ROI of 2cm diameter was segmented by aligning the centre of the ROI at the centre of the insert. The Matlab package was used to segment and extract image matrices that were used to perform hand GLCM calculations. A kV Cone Beam Computed Tomography (kV CBCT) acquired cervical cancer data-set was used to establish the robust radiomic texture features response to radiotherapy treatment. The kV CBCT images were acquired first day and weekly during the 25 treatment fractions. Results: Five first order statistic radiomic features and six grey level co-occurrence matrix features were identified in the experimental test and mathematical manual calculations tests to vary with coefficients of variance of less than or equal to 10 % when the slice thickness was varied. Most of the radiomic texture features were weak and unstable (coefficients of variance above 10%) at very small slice thickness (≥2.5 mm) and robust at medium (≥2.5 mm) to large slice thickness (3.75 mm and 5 mm) (coefficients of variance ≤ 10 %). The above was attributed to an averaging effect (image smoothening) on the images when the slice thickness of image acquisition is increased. The image noise was observed to be less in large slice thickness when compared to noise at small slice thickness. Radiomics features were independent and stable to the tube potential at greater than 100 kV. At high tube potential the radiation attenuated signal detected at the CT detector was higher cancelling the noise effects. The robustness of these radiomic features depended on the material comprising the insert analysed. The extent of mAs dependence observed for the dense cork and plaster resin materials inserts was low compared to the dependence on the solid acrylic material insert. All the other phantom inserts (rubber particles, natural cork and the 3 acrylonitrile butadiene styrene plastic) data plots showed smaller variations around the central axis (zero feature value) of the skewness, uniformity, entropy and kurtosis features graphs. Irrespective of the mAs changes, the radiomic texture feature values obtained from all of the ABS materials inserts, rubber particles and natural cork inserts were consistently smaller, closer to zero. A general decrease in image noise as the mAs of image acquisition was increased in images of uniform or relatively uniform material was also observed. The patient tumour analysis showed some radiomic texture features response to radiotherapy treatment. This was shown by the changes observed on the inverse difference, inverse difference moment, entropy and difference variance texture features. The texture features had their values decrease from start of treatment (first fraction) to the last treatment fraction. The decrease was not smooth along the treatment period, there were some anomalies on the trends. This decrease was ascribed to the change in the heterogeneity of the tissues within the treatment region of interest evaluated. Conclusion: Overall, using theoretical analysis and a practical approach, robust radiomic features that were independent of the CT scan parameters were observed. The experimental approach showed that the phantom insert materials had influence on radiomic texture feature values obtained in investigations. Radiomic texture features demonstrated that tumours had a variation of heterogeneity between them. The observation agrees with other clinical studies that showed that tumours exhibit some extensive genetic and phenotypic variations. Radiomic texture features can be utilised to depict tumour texture changes along the treatment timeline as shown in this study. A great challenge would be to associate the radiomic texture feature changes to the clinical biological changes. For future robust radiomic feature studies, the use of phantoms with tissue like materials was proposed.
Description
Keywords
Dissertation (M.Med.Sc. (Medical Physics))--University of the Free State, 2019, Radiomic texture features, Computed tomography, Tumour, Imaging parameters, Slice thickness, Tube potential difference, Tube current, Software, Robust texture features, Credence Cartridge Radiomics (CCR) phantom