Potential of MRI-based radiomics in non-invasive pre-operative assessment of lymphovascular invasion in cervical cancer

Objective. To develop and validate a magnetic resonance imaging (MRI)-based prognostic model for pre-operative assessment of lymphovascular invasion (LVI) in cervical cancer (CC). Patients and methods. Clinical data from 109 patients aged 23–82 years with stage IA2-IIA2 CC were retrospectively analyzed. All patients were randomly divided into the study group (n = 81) and the test (n = 28) group. At the pre-operative stage, all patients underwent pelvic MRI. The most significant MRI signs were identified and radiomic models were developed using the LASSO linear regression algorithm based on data from the study group. The radiomic nomogram was developed considering the rad-score (based on the evaluation of radiomic parameters) and clinical risk factors for LVI (age, maximum linear size, degree of tumor differentiation). The informativeness of models was assessed using ROC analysis and decision curve analysis. Results. The radiomic nomogram model developed with 15 radiomic parameters and clinical features showed high informativeness in both patient cohorts: in the study group, accuracy was 77.8%, sensitivity 77.8%, specificity 77.8%, and in the test group: 82.1%, 80.0%, 84.5%, respectively. ROC analysis demonstrated no significant differences in the efficacy of prognostic model between the study and test groups (AUC 0.84 and 0.83, respectively). Conclusion. The developed comprehensive prognostic model based on the combination of radiomic parameters and clinical data has high diagnostic efficacy in the pre-operative prediction of LVI in CC, including validation in the test group. Key words: magnetic resonance imaging, oncology, cervical cancer, radiomic nomogram