Investigating the Reproducibility of Radiomics Features Extracted From Ultrasound Images As Diagnostic Biomarkers in Patients With Hepatocellular Carcinoma; [بررسی تکرارپذیري فیچرهاي رادیومیکس استخراج شده از تصاویر اولتراسوند بهعنوان بیومارکرهاي تشخیصی در بیماران مبتلا به کارسینوم سلولهاي کبدي ]



Soleymani Y¹ ; Jahanshahi AR² ; Rezaeejam H³ ; Khezerloo D⁴ Show Affiliations
Source: Tehran University Medical Journal Published:2023

Abstract

Background: Radiomics is a noninvasive method that reveals information from medical images that are not recognizable by the naked eye. Radiomics has shown a high potential in the accurate diagnosis and prognosis of liver lesions in ultrasound images. Despite this high potential, changes in imaging parameters affect the reproducibility of ultrasound radiomics results. Therefore, the present study aims to investigate the reproducibility of the radiomics features extracted from the images of patients with hepatocellular carcinoma under changes in ultrasound scan parameters. Methods: This was a cross-sectional study conducted from July 2020 to July 2021 in the radiology department of Tabriz Paramedical Faculty. The images of 20 patients with hepatocellular carcinoma were obtained from the Cancer Imaging Archive database. These images were taken under different imaging conditions and parameters. The areas related to the lesion were manually extracted from the images with software tools. Then, in order to radiomics analysis, different radiomics features, including 24 gray level co-occurrence matrix (GLCM) and 16 gray level run length matrix (GLRLM), were extracted from the images. Then, using the coefficient of variation (CV%) and intraclass correlation coefficient (ICC) statistical tests, the reproducibility of radiomics features under changes in scan parameters was investigated. The values of ICC≥0.90 and CV<20% were considered reproducible in this study. Results: Among the 40 features extracted from ultrasound images, eight showed high reproducibility in both CV% and ICC tests. These features were joint entropy, Idmn, Imc2, correlation, MCC, sum entropy, gray level non-uniformity normalized, and run entropy in which the two features, Idmn and gray level non-uniformity normalized, showed the highest (CV%=0.24) and the lowest (CV%=14.90) stability against the changes of ultrasound scan parameters, respectively. The average ICC value of these features was obtained at 0.977. Conclusion: Despite the high potential of radiomics in diagnosing liver lesions, changes in imaging parameters directly affect the reproducibility of results. However, some radiomics features still show high stability and reproducibility under changes in imaging parameters. Copyright © 2023 Soleymani et al.