The Potential of Radiomics in the Assessment of Intestinal Fibrosis in Crohn’s Disease

We read with great interest the article by Li et al1Li X. et al.Gastroenterology. 2021; 160: 2303-2316.e11Abstract Full Text Full Text PDF PubMed Scopus (12) Google Scholar published in the June 2021 issue of Gastroenterology. This study aimed to develop and validate a radiomic model based on computed tomographic enterography (CTE) images for grading (none-mild vs moderate-severe) intestinal fibrosis in patients with Crohn’s disease (CD). The results indicated that in the external validation sets the radiomic model achieved areas under the receiver operating characteristic curve (AUC) from 0.724 to 0.816, which significantly outperformed radiologist-interpreted imaging signs (AUCs <0.600) in the evaluation of the severity of intestinal fibrosis in CD. This retrospective small study has some merits that deserve to be noted and provides insights into future radiomic work to improve the study quality. First, while most radiomic studies focus on cancer, this study brings the radiomic approach to the field of inflammatory bowel disease. Second, the authors estimated the required sample size to reach an AUC of at least 0.800. The sample size calculation has been rarely performed in previous radiomic papers.2Sollini M. et al.Eur J Nucl Med Mol Imaging. 2019; 46: 2656-2672Crossref PubMed Scopus (93) Google Scholar Third, this study performed subgroup analysis by factors already known to test the robustness of the radiomic model, including CD location, CT scanners, and bowel strictures with or without penetrating diseases. The results showed that the performance of the radiomic model was not obviously fluctuant. This action may inspire us to conduct a subgroup analysis when planning an image-mining study to prove the reliability of a predictive or diagnostic model. Fourth, this study compared the diagnostic performance of the radiomic model with radiologist-reported imaging findings. It helps to determine whether radiomics improves current clinical practice. Finally, the design of this study adhered to a standard radiomic workflow, including image process, feature extraction and selection, modeling, and clinical usefulness evaluation. Despite the promising results, we are concerned about key aspects of the analysis. The image preprocess can reduce the density variations among different CT scanners and thus is a crucial step before feature extraction, yet this important step was not included in this study. Considering the small sample size of this study, data augmentation can be properly used to avoid model overfitting. We appreciate the stratification analysis to prove the stability and robustness of the radiomic model, but it is post hoc and the authors should take into account these factors during the process of feature selection. For example, the authors might determine the effect of different CT scanners and acquisition parameters on the robustness of extracted radiomic features, and include only features that did not show significant differences due to machine characterization and acquisition parameters. To further enhance the discrimination power of the radiomic model, it may be beneficial to incorporate the clinical data and radiologic findings to construct a combined model. Different types of medical data sources in practical clinical scenarios are encouraged to be combined to create a reliable diagnostic model. Regarding the model reading compared with human reading, because the radiologist-performed visual interpretation, instead of the direct diagnosis by these observers, was evaluated by means of logistic regression analysis, it may be not appropriate to conclude that the diagnostic performance of the radiomic model was superior to that of the radiologists. The diagnosis by radiologists is based on a comprehensive observation and is not limited to the 3 key imaging findings reported in the Methods section. The 95% confidence intervals for the AUCs were very wide in the test data sets, suggesting that the diagnostic performance was unstable and may be changed in a larger patient population. Given the small sample size and methodology limitations, this study may overstate its conclusion and we should interpret the findings with caution. Although this study describes a relationship between the severity of intestinal fibrosis and radiomic features, the biological meaning of these features needs to be investigated in the future, which is also an unresolved issue in the radiomics field.3Tomaszewski M.R. Radiology. 2021; 299: E256Crossref PubMed Google Scholar The workflow of radiomics is complex, and each step can affect the results of a radiomic model. Therefore, standardization and rigorous quality control are needed for radiomics to succeed. The design and report of a radiomic study may follow the rigorous radiomics quality scoring and transparent reporting of the clinical radiomics study methods proposed by Lambin et al.4Lambin P. et al.Nat Rev Clin Oncol. 2017; 14: 749-762Crossref PubMed Scopus (1741) Google Scholar The development of a reliable group of radiomic biomarkers and criteria for evaluating those features in multiple prospective cohorts are challenging but can accelerate the clinical application of radiomics.5Halligan S. et al.Eur Radiol. 2021 May 18; https://doi.org/10.1007/s00330-021-07971-1Crossref Scopus (7) Google Scholar Further work is warranted to determine whether the potential demonstrated in this study can be broadly validated and to develop an available online web-based tool for clinical use. Development and Validation of a Novel Computed-Tomography Enterography Radiomic Approach for Characterization of Intestinal Fibrosis in Crohn’s DiseaseGastroenterologyVol. 160Issue 7PreviewNo reliable method for evaluating intestinal fibrosis in Crohn’s disease (CD) exists; therefore, we developed a computed-tomography enterography (CTE)–based radiomic model (RM) for characterizing intestinal fibrosis in CD. Full-Text PDF ReplyGastroenterologyVol. 161Issue 6PreviewWe thank Zhang et al1 for their comments on our article2 and appreciate the opportunity to reply to these 5 comments. Full-Text PDF