Our previous experience confirmed that MRI staging-based radiomics models provide a relevant predictive tool to identify the oncological behavior of pCR after nCRT [24]. Despite the important effort made in terms of treatment response prediction, few experiences reported the relation between radiomics predictors and early distant recurrence [25, 26]. Our results show that adding radiomics features to the clinical model can better predict SDM performance in rectal cancer patients, increasing the AUC from 0.83 to 0.94, with high sensitivity and specificity in the validation cohort. The high specificity indicates that the model is reliable and can eliminate more false positive and false negative patients. We developed a clinical-radiomics nomogram as an individual and visualization tool to provide an estimated probability of SDM for newly diagnosed rectal cancer patients. Decision curve analysis was used to determine its clinical benefit. Radiomics, integrating many high-dimensional imaging features used to quantify tumor heterogeneity, could facilitate oncologic diagnosis and prognosis prediction. Huang et al [22] revealed that radiomics signature including 24 selected features could help predict LN-positive patients with a C-index of 0.773 in the validation cohort and the proposed clinical-radiomics nomogram was useful for predicting LN involvement. In our study, radiomics features were obtained using high-resolution axial T2WI and DWI, which are the most critical sequences for evaluating primary rectal cancer[27]. Functional imaging techniques such as DWI can analyze lesions from the perspective of the microenvironment and molecular pathology, to quantitatively study many small lesions that cannot be quantified and identified by the naked eye. Our proposed clinical-radiomics predictive model confirms the feasibility of imaging analysis based on T2WI and DWI, providing a potentially effective and easy-to-use model in clinical practice. Using our clinical-radiomics nomogram, an estimated probability of SDM could be calculated after referring to the selected T2WI-based radiomics features, DWI-based radiomics features as well as other clinical information.
Our results show that the radiomics nomogram provides predictive information about SDM in primary rectal cancer. Contrast-enhanced CT, MRI, and PET-CT are common imaging examinations for the diagnosis of SLM in RC preoperatively. However, the sensitivity and accuracy of these imaging techniques are not satisfactory[28–30]. According to one meta-analysis, the detection sensitivity of colorectal LM in contrast-enhanced CT, routine MRI, and FDG PET-CT were 63%-80%, 76%-85.7%, and 51%-90%, respectively[31]. Many studies have shown that MRI had a higher accuracy compared to CT in diagnosing SLM of rectal cancer patients[32–34], and recent consensus guidelines from the radiologic community recommend MRI for the preoperative evaluation of SLM [35, 36]. The other studies had shown that some adverse features found on rectal MRI identified patients with rectal cancer at higher risk of distant metastasis[37–39]. Therefore, we developed a radiomics nomogram based on multi-sequence MRI to predict the risk of SDM.
Therefore, screening for high-risk predictors would improve the probability of early detection of SDM in RC patients. Typically, the clinicopathological predictors of SDM in RC patients include the histological type, pathological grade, depth of tumor invasion, lymph node status, vascular invasion, and tumor markers[40]. However, some of these predictors can only be obtained postoperatively, and hence, inappropriate to guide preoperative treatment. Other studies have demonstrated that some features of rectal MRI, such as extramural vascular invasion, higher T stage, and regional lymph node metastasis are potential predictors[38, 39]. However, these image features are subjective and qualitative, lacking quantitative assessment. In recent years, radiomics has been regarded as an advanced tool for evaluating tumor heterogeneity in tumor diagnosis and prognosis prediction. In this study, factors such as radiomics features, CEA, and CA19-9 levels were incorporated into multivariate logistic regression to build a prediction model and a nomogram, and the research results are promising. Therefore, our analysis indicates that the radiomics nomogram combined with tumor markers was superior to the radiomics signature alone. It exhibited a high predictive performance for SDM in RC patients, and the AUC improved from 0.82 to 0.94. Moreover, the results were better than those reported in a previous study on a per-patient basis, wherein the AUC was 0.92 and 0.88 (MRI readers), 0.80 and 0.82 (CT readers), and an AUC of 0.83 and 0.84 (PET-CT readers)[41]. In this study, we constructed a primary RC-based radiomics nomogram from high-resolution T2WI and DWI of the recta to predict SDM. On the training set, the AUC, accuracy, sensitivity, specificity, and 95%CI of the nomogram model were 0.93, 0.85, 0.85, 0.86, and 0.89–0.96, respectively. The AUC, accuracy, sensitivity, specificity, and 95% CI of the nomogram model were 0.94, 0.89, 0.89, and 0.79 ~ 0.97 respectively. Therefore, based on clinical risk factors and radiomics characteristics, the proposed nomogram may be a valuable predictive tool for SDM in RC patients. It can be easily used to identify patients who need further whole-body imaging.
Radiomics features are composed of multiple features, which is of great significance for selecting the optimal feature collection through dimensionality reduction. In this study, we used the LASSO method for feature screening and built a radiomics model based on a support vector machine (SVM). In our study, the method of noose dimension reduction is chosen to increase the stability of the nomogram and to carry out the overall analysis. The nomogram we constructed is easier for clinicians to use because it can derive information from T2WI, and DWI, and includes clinical risk factors. The radiomics nomogram is an auxiliary tool that can be used to identify and follow those patients with rectal cancer. When the group was divided into a low-risk group and a high-risk group, the high-risk group had a higher probability of developing SDM. Therefore, in a certain sense, the radiomics nomogram can be used as an accurate and reliable detection tool for SDM in rectal cancer patients. It is quick and easy to perform and helps to determine which patients will benefit from further imaging of distant metastases.
However, the current study still has some limitations. First of all, the sample size based on single-institution retrospective analysis is relatively small, and selection bias is inevitable. Second, this study lacks external validation, so a large multicenter trial is needed to improve the generalizability of the results.
In summary, we developed a clinical imaging nomogram by combining tumor markers with imaging features to accurately predict the presence of SDM in RC patients. This visualization tool will detect the probability of SDM and help doctors make clinical decisions.