Leveraging Artificial Intelligence and Radiomics for Improved Nasopharyngeal Carcinoma Prognostication

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Nasopharyngeal carcinoma (NPC) often presents at advanced stages, making accurate prognostication difficult.
• Current anatomical staging lacks the precision to differentiate patient prognoses effectively.
• This study explores radiomics for improved prediction of locoregional recurrence (LRR) and overall survival in NPC.

Purpose of the Study:

• To investigate the utility of radiomics in enhancing the prediction of LRR and overall survival in NPC.
• To develop and validate models combining clinical data and radiomic features for NPC prognostication.

Main Methods:

• Radiomic features were extracted from CT scans of 294 NPC patients.
• A feature selection process identified six key radiomic features.
• Models were trained using clinical data, radiomic features, and their combination to predict 2-year LRR.

Main Results:

• The combined model (clinical data + radiomics) achieved the highest AUC (0.76) for predicting 2-year LRR.
• Clinical data alone yielded an AUC of 0.56, and radiomics alone yielded 0.57.
• Risk stratification using the combined model significantly impacted LRR-free and overall survival (p < 0.01).
• Key radiomic features included tumor size, intensity distribution, and textural patterns.

Conclusions:

• Radiomics shows significant promise for improving NPC risk stratification and enabling personalized treatment.
• Tumor size, identified as a key radiomic feature, warrants reconsideration as a prognostic criterion.
• Further prospective studies are necessary to validate these radiomics-based findings.