Computed tomography radiomics signature via machine learning predicts RRM2 and overall survival in hepatocellular carcinoma

Qian Li,Xiawei Long,Yan Lin,Rong Liang,Yongqiang Li,Lianying Ge

DOI: https://doi.org/10.21037/jgo-23-460

2023-06-30

Journal of Gastrointestinal Oncology

Abstract:Background: Radiomics can be used to noninvasively predict molecular markers to address the clinical dilemma that some patients cannot accept invasive procedures. This research evaluated the prognostic significance of the expression level of ribonucleotide reductase regulatory subunit M2 (<i>RRM2</i>) in individuals with hepatocellular carcinoma (HCC) and established a radiomics model for predicting the <i>RRM2</i> expression level.Methods: Genomic data for HCC patients and corresponding computed tomography (CT) images were accessed at The Cancer Genome Atlas (TCGA) and The Cancer Imaging Archive (TCIA), which were utilized for prognosis analysis, radiomic feature extraction and model construction, respectively. The maximum relevance minimum redundancy algorithm (mRMR) and recursive feature elimination (RFE) were used for feature selection. Following feature extraction, a logistic regression algorithm was fitted to establish a dichotomous model that predicts <i>RRM2</i> gene expression. Establishment of the radiomics nomogram was carried out using the Cox regression model. Receiver operating characteristic (ROC) curve analysis was employed to assess the model performance. Clinical utility was determined by decision curve analysis (DCA).Results: High <i>RRM2</i> expression acted as a risk factor for overall survival (OS) [hazard ratio (HR) =2.083, P&lt;0.001] and was implicated in regulation of the immune response. Four optimal radiomics features were selected for prediction of <i>RRM2</i> expression. A predictive nomogram was established using the clinical variables and radiomics score (RS), and the areas under the ROC curve (AUCs) of the time-dependent ROC curve of the model were 0.836, 0.757, and 0.729 for the 1-, 3-, and 5-year periods, respectively. DCA confirmed that the nomogram had good clinical usefulness.Conclusions: The <i>RRM2</i> expression level in HCC can considerably affect prognosis of these patients. Expression levels of <i>RRM2</i> and prognosis of HCC individuals can be predicted through radiomics features by utilizing CT scan data.

oncology,gastroenterology & hepatology