Radiomics is a promising analysis tool consisting in extracting quantitative information from medical images. However, the sensitivity of the extracted radiomics features to variations in used image acquisition and reconstruction parameters hinders the validity of radiomics-assisted models. In this study we investigated the repeatability and robustness of radiomics features using a new phantom designed specifically for radiomics studies called Radiomik [1]. 32 CT protocols were used to acquire the Catphan-500 phantom images, 10 of these were chosen as the starting point to acquire Radiomik’s scans. Two scanners, different dose levels and percentages of Iterative Reconstruction algorithms were considered. The repeatability of the features was tested by evaluating the Coefficient of variation (CV) using different insert’s ROI placements, volumes and considering three repetitions of the same protocol. To assess features robustness the Intraclass Correlation Coefficient (ICC) was calculated. The repeatability tests showed that some features have a strong dependence on the positioning and volume of the ROIs. In the robustness analysis a higher ICC was achieved for inserts more homogeneous and with higher filling percentages while worse results were obtained for inserts with lower filling percentages and with a greater number of dark pixels. This study highlights that, while a harmonization based on image quality could improve features robustness as explored in a previous work [? ], it is necessary to also design a more detailed scheme for both image acquisition and features extraction to allow multi-centric studies and thus to enhance the validity of radiomics models.
SC10.04 EVALUATION OF REPEATABILITY AND ROBUSTNESS OF CT-DERIVED RADIOMIC FEATURES USING A CUSTOM PHANTOM
Tenerani, M. I.
Primo
;Imbriani, M.;Lizzi, F.;Quattrocchi, M.;Scapicchio, C.;Zafaranchi, A.;Fantacci, M. E.Ultimo
2024-01-01
Abstract
Radiomics is a promising analysis tool consisting in extracting quantitative information from medical images. However, the sensitivity of the extracted radiomics features to variations in used image acquisition and reconstruction parameters hinders the validity of radiomics-assisted models. In this study we investigated the repeatability and robustness of radiomics features using a new phantom designed specifically for radiomics studies called Radiomik [1]. 32 CT protocols were used to acquire the Catphan-500 phantom images, 10 of these were chosen as the starting point to acquire Radiomik’s scans. Two scanners, different dose levels and percentages of Iterative Reconstruction algorithms were considered. The repeatability of the features was tested by evaluating the Coefficient of variation (CV) using different insert’s ROI placements, volumes and considering three repetitions of the same protocol. To assess features robustness the Intraclass Correlation Coefficient (ICC) was calculated. The repeatability tests showed that some features have a strong dependence on the positioning and volume of the ROIs. In the robustness analysis a higher ICC was achieved for inserts more homogeneous and with higher filling percentages while worse results were obtained for inserts with lower filling percentages and with a greater number of dark pixels. This study highlights that, while a harmonization based on image quality could improve features robustness as explored in a previous work [? ], it is necessary to also design a more detailed scheme for both image acquisition and features extraction to allow multi-centric studies and thus to enhance the validity of radiomics models.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
