Research
Radiographic Image Radiomics Feature Reproducibility: A Preliminary Study on the Impact of Field Size

https://doi.org/10.1016/j.jmir.2019.11.006Get rights and content

Abstract

Rationale and Objectives

Radiomics is an approach to quantifying diseases. Recently, several studies have indicated that radiomics features are vulnerable against imaging parameters. The aim of this study is to assess how radiomics features change with radiographic field sizes, positions in the field size, and mAs.

Materials and Methods

A large and small wood phantom and a cotton phantom were prepared and imaged in different field sizes, mAs, and placement in the radiographic field size. A region of interest was drawn on the image features, and twenty two features were extracted. Radiomics feature reproducibility was obtained based on coefficient of variation, Bland-Altman analysis, and intraclass correlation coefficient. Features with coefficient of variation ≤ 5%, intraclass correlation coefficient ≤ 90%, and 1% ≤ U/LRL ≤30% were introduced as robust features. U/LRL is upper/lower reproducibility limits in Bland-Altman.

Results

For all field sizes and all phantoms, features including Difference Variance, Inverse Different Moment, Fraction, Long Run Emphasis, Run Length Non Uniformity, and Short Run Emphasis were found as highly reproducible features. For change in the position of field size, Fraction was the most reproducible in all field sizes and all phantoms. On the mAs change, we found that feature, Short Run Emphasis field 15 × 15 for small wood phantom, and Correlation in all field sizes for Cotton are the most reproducible features.

Conclusion

We demonstrated that radiomics features are strongly vulnerable against radiographic field size, positions in the radiation field, mAs, and phantom materials, and reproducibility analyses should be performed before each radiomics study. Moreover, these changing parameters should be considered, and their effects should be minimized in future radiomics studies.

Résumé

Justification et objectifs

La radiomique est une nouvelle approche de quantification de la maladie. Récemment, plusieurs études ont indiqué que les caractéristiques de radiomique étaient vulnérables Le but de cette étude est d’évaluer de quelle façon les caractéristiques de radiomique changent avec la taille du champ radiographique, les positions dans la taille du champ et les mA.

Matériel et méthodologie

Deux fantômes de bois, un grand et un petit, ainsi qu'un fantôme de coton ont été radiographiés avec des champs de différentes tailles, à différents mA et dans des positions différentes dans le champ. Une région d'intérêt a été dessinée sur les caractéristiques des images et 22 caractéristiques ont été extraites. La reproductibilité des caractéristiques de radiomique a été obtenue à partir du coefficient de variation (COV), d'un graphique de Bland-Altman (BA) et d'un coefficient de corrélation intraclasse (ICC). Les caractéristiques correspondant à COV ≤ 5 %, ICC ≤ 90 % et 1 % ≤ U/LRL ≤ 30 % ont été retenues comme caractéristiques robustes. U/LRL est la limite de reproductibilité supérieure/inférieure dans BA.

Résultats

Pour toutes les tailles de champ et tous les fantômes, les caractéristiques Difference Variance, Inverse Different Moment, Fraction, Long Run Emphasis, Run Length Non Uniformity et Short Run Emphasis se sont avérées hautement reproductibles. Pour les changements de position dans la taille du champ, la caractéristique Fraction était la plus reproductible pour toutes les tailles de champ et tous les fantômes. Pour le changement de mA, nous avons constaté que les caractéristiques Short Run Emphasis pour un champ de 15 × 15 avec le petit fantôme de bois et Correlation pour le fantôme de coton dans toutes les tailles de champ étaient les caractéristiques présentant la plus grande reproductibilité.

Conclusion

Nous avons démontré que les caractéristiques de radiomique étaient fortement vulnérables selon les facteurs de taille du champ radiographique, de position dans le champ, de mA et de construction des fantômes, et que des analyses de reproductibilité devraient être effectuées avant chaque étude de radiomique. De plus, ces paramètres changeants devraient être pris en compte et leurs effets minimisés dans les études de radiomique futures.

Introduction

Medical imaging plays a critical role in the management of different patients in terms of disease detection, diagnosis, treatment planning, and follow-up [1]. New advances in medical imaging have given medical specialists the ability to manage diseases using both qualitative and quantitative approaches [2]. In the field of oncology, imaging has been found in a wide range of applications and utilized more in cancer management as diagnostic, prognostic, and predictive markers [[3], [4], [5]]. Recently, quantitative radiomics analysis, as a newly accepted branch of advanced image processing, has opened a new horizon of biomarker discovery in the field of cancer and many different types of disorders [6,7].

As defined by previous studies, “radiomics is the high-throughput extraction of large amounts of image features from radiographic images creating a high-dimensional data set followed by data mining for potentially improved decision support” [8]. Radiomics is a multistep process involving the acquisition of high-quality diagnostic or planning medical images, definition of the tumor with manual or automated segmentation methods, extraction of quantitative radiomics features from the segmented regions, selection of most informative features, feature analysis for their relationship with clinical or biological data, and finally building diagnostic, prognostic, and predictive models by using these radiomics features [8].

Radiomics features are mathematical issues that are a measure of spatial relationship and distribution of intensity, heterogeneity patterns, shape, size, and relations of segmented regions (tumor) with surrounding regions (other tissues) [9]. These quantitative data, which are newly accepted biomarkers, are well-defined descriptors of tumor characteristics, biology, and behaviors against therapeutic approaches [10]. Although these imaging features have been issued as cost-beneficial, easy-to-use, and noninvasive, several challenges remain regarding the use of these markers [[11], [12], [13], [14]]. Based on imaging biomarker recommendations, radiomics features have remained unchanged and robust against any changes in each radiomics process [15]. Previous studies have indicated that many image features are vulnerable against different image acquisition, reconstruction, and segmentation in all imaging modalities including radiology, computed tomography, magnetic resonance imaging, and positron emission tomography [14,[16], [17], [18], [19]].

Although radiomics is used more for oncological data and in three dimensional (3D) features, studies have demonstrated the wide application of imaging features in 2D and for conventional radiographic imaging. Radiographic radiomics feature analysis is an applicable approach to quantify several diseases and also for computer-aided detection or diagnosis. Computer-aided detection or diagnosis has been proven to be an effective approach to assist radiologists in improving diagnostic accuracy [20,21].

To the best of our knowledge, there are few studies on the radiographic radiomics feature reproducibility. In our previous study, we showed the effect of radiographic parameters including kVp, mAs, source to film distance, filtration, and tube angle variations on feature values [12]. In the present study, we aimed to assess how radiomics features change with radiographic field sizes and positions in the field size. In this work, we assessed these changes on phantoms made with different materials and also imaged with different mAs.

Section snippets

Materials and Methods

Figure 1 shows the flowchart of the study.

Field Size Change

Our results on COV, ICC, and BA for radiomics feature reproducibility against changes in field size for all phantoms, CO and big and small woods, are depicted in Figure 3.

In regard to COV for all phantoms, six features including DifEntrp and InvDfMom from gray-level co-occurrence matrix (GLCM) feature set and Fraction, LngREmph, RLNonUni, and ShrtREmp from gray-level run length matrix (GLRLM) feature set are found as highly reproducible features. For CO and big and small woods, there was a

Discussion

Radiomics is an emerging field of medical imaging that has the potential to improve personalized medicine by using mineable imaging data, noninvasively [12,14,[22], [23], [24], [25], [26], [27]]. Based on previous research, radiomics has played a significant role in the development of new feasible biomarkers in radiation oncology. These biomarkers will be used clinically, if they have the main requirements such as repeatability and reproducibility [17].

In the present study, the impact of

Conclusion

In conclusion, we demonstrated that radiomics features are strongly vulnerable against radiographic field size, positions in the radiation field, mAs, and phantom materials. The results of this study could be taken into account when a quantitative imaging is needed for patient outcome analysis and disease detection and management.

References (33)

View full text