Enhanced CT Based Radiomics in Predicting the Recurrence for Patients with Extranodal Natural Killer T-Cell Lymphoma

ABSTRACT

Background and Purpose:Extranodal Natural Killer/T-Cell Lymphoma (ENKTL) is a highly aggressive non-Hodgkin lymphoma, primarily originating from NK cells, with some cases arising from T-cell lineage. The primary site of ENKTL lesions is the nasal region, examined using methods such as nasal endoscopy, CT, MRI, and PET/CT. Radiomics, an emerging field of translational research, extracts quantitative features from medical images and integrates them with conventional prognostic markers like clinical staging, molecular biomarkers, and pathological characteristics. This study aims to construct a radiomics model based on contrast-enhanced CT to predict the prognosis of ENKTL and evaluate its predictive value.

Materials and Methods: A retrospective study included 115 ENKTL patients diagnosed and treated in our hospital from January 2011 to January 2021. All patients underwent enhanced CT scans, and their clinical data and follow-up information were collected. Regions of interest (ROI) were delineated, and CT radiomics features were extracted. The data were divided into training and validation groups (6:4 ratio). The least absolute shrinkage and selection operator-Cox regression model was used to select predictive radiomics features, which were then used to create a radiomics scoring model. This model's performance was evaluated using time-dependent ROC curve analysis and the AUC. The optimal radiomics score threshold was determined using X-tile software, dividing patients into high and low score groups. Univariate and multivariate Cox regression analyses evaluated the association of clinical factors with relapse-free survival. Based on multivariate Cox regression results, clinical and radiomics nomograms were established to predict 3-year and 5-year relapse-free survival. The radiomics nomogram incorporated radiomics scores and clinical risk factors, while the clinical nomogram included only independent clinical risk factors. The clinical validity of the radiomics nomogram was confirmed by comparing the net benefit relative to threshold probabilities on the decision curve.

Results: A total of 851 radiomics features were extracted, and 3 key features were selected to establish a radiomics scoring model. The waterfall chart indicated a positive correlation between the radiomics score and the recurrence status. The radiomics scoring model demonstrated strong discriminatory performance, as suggested by the AUC values of the ROC curve. Kaplan-Meier curve analysis showed a significant correlation between radiomics score and recurrence-free survival. Patients in the low-scoring group had shorter total recurrence times and higher recurrence rates compared to the high-scoring group. The C-index of the radiomics nomograms in both cohorts indicated that the radiomics nomogram had better predictive performance than the clinical nomogram. Decision curve analysis further confirmed that the radiomics nomogram provided greater predictive value than the traditional clinical nomogram.

Conclusion: This study shows that the radiomics model based on enhanced CT has good value in predicting the prognosis of ENKTL patients.

No relevant conflicts of interest to declare.