摘要
See the article by Beig et al, pp. 251–263. “Radiogenomics” is a term being increasingly used for the field of science combining imaging features with clinical and genomic data to develop better prognostic and predictive models for various oncologic diseases. Under this broader use of the term “radiogenomics” also falls the field of “radiomics” which incorporates high-throughput quantitative image analysis and use of computer-generated image features going beyond the traditional radiologist-driven subjective assessment of radiological scans. Both these fields have opened up avenues to interrogate oncological diseases such as gliomas from an angle not previously appreciated, and the number of published research articles on this topic have grown exponentially in the last decade. However, the field of radiomics lacks standardized evaluation of both the scientific integrity, and more importantly, the clinical relevance of the majority of these publications. Rigorous evaluation criteria and reporting guidelines need to be established for radiomics-driven approaches to be employed and impactful in actual clinical practice.1 A radiomics quality score (RQS) has been proposed to assess the quality of clinical evidence in these publications which is based on a number of parameters including study design, external validation, reproducibility, clinical utility, and open science.1 A recent systematic review has shown that the quality of radiomics studies in neuro-oncology is rather dismal.2 In this issue of Neuro-Oncology, Beig et al. have explored these radiogenomic concepts and identified gender-based radiomics differences in glioblastomas (GBMs) that correlate with patient survival.3 Study by Beig et al. scored fairly well (RQS = 21 out of 36), with a great in-depth study design, validation, and potential clinical utility.3 Importantly, the authors developed and validated a radiomics risk score within each gender as well as for the entire population. Although female sex previously has been associated with improved survival in GBM,4,5 potential imbalance in important prognostic or predictive glioma molecular factors between sexes could confound survival analyses, and subsequently, any observed imaging associations correlating with sex and prognosis. For example, MGMT methylation, which predicts for response to standard chemoradiotherapy in GBM, has been found to be more prevalent in females.4,6,7 In the study by Beig et al., groups with high or low gender-specific risk based on radiomics risk scores were associated with survival within each gender, controlling for potential prognostic confounders.3 In addition, this design enabled Beig et al. to identify candidate biological processes associated with not only overall radiomic risk scores but specific individual radiomic features within each gender as well. For example, in males, the high-risk radiomic risk score was associated with higher expression of genes associated with angiogenesis and cell adhesion, whereas in females, the gender-specific radiomic risk score was associated with pathways of immune response.3 Interestingly, high expression of the Laws energy radiomic feature from peri-tumoral areas was positively correlated with angiogenesis pathways in males, highlighting the potential for radiomics to reveal pathogenic biological mechanisms. These data provide pathways for interrogating mechanisms not only driving radiomic imaging features but also contribute to the emerging data on the mechanisms underpinning the different therapeutic responses to the same therapy in females with GBM compared to males.3,8 One of the major shortcomings is lack of a prospective arm and hence not providing the highest level of evidence supporting clinical validity and usefulness. Radiomic studies usually lead to creation of high dimensionality of extracted features and hence, the risk of overfitting if feature selection or reduction is not employed. Another major issue with some of these heavy data-driven approaches is independent and external validation and hence reproducibility of the results. Beig et al. addressed these common shortcomings of retrospective studies well. Another major reason radiomics studies will face uphill battle in their clinical implementation is the lack of “explainers” for radiomic features in the common vernacular that clinical radiologists and neuro-oncologists understand.9 For example, Laws energy features (capturing spots and ripples-like patterns or detecting levels and edges) have no relevance to commonly practiced radiological lexicon and hence, additional work to correlate these radiomics-derived features with commonly observed imaging features or development of terminology for these radiomic features in standard-of-care medical imaging may be required prior to universal acceptance in the clinic. In summary, Beig et al. have demonstrated improvements not just in the level of evidence provided by a well-conducted research, but also have shown that radiomics could explore biological pathways and gender-based differences in GBMs beyond what could be achieved with conventional medical imaging assessment. Further prospective validation for radiomics and radiogenomics research will help establish these tools in clinical practice. Conflict of interest statement. A.S.C. is an employee of Mirati Therapeutics. Authorship statement. R.J. prepared the manuscript; A.S.C. revised the manuscript critically for important intellectual content.