Researchers have developed and validated an AI-based deep learning system to better predict outcomes in patients with oropharyngeal cancer (OPC) using CT scans taken before treatment. The tool focuses on extranodal extension (ENE), a condition in which cancer spreads outside the lymph node. ENE is usually confirmed only after surgery, but this new AI model can predict it from initial imaging.
The study found that the number of lymph nodes with AI-predicted ENE is strongly linked to survival. Patients with two or more ENE-positive nodes had a much higher risk of cancer spreading to distant organs and a higher risk of death. Adding the AI-predicted ENE count to the current AJCC 8th Edition staging system significantly improved the accuracy of predicting 3-year survival and distant metastasis. The benefit was especially strong in HPV-negative patients, who generally have worse outcomes.
The AI system works in two steps. First, a 3D neural network automatically identifies and measures suspicious lymph nodes on CT scans. Then, a model called DualNet analyzes those nodes to predict the likelihood of ENE, performing better than experienced radiologists. This tool is clinically important because many OPC patients receive radiation or chemotherapy instead of surgery and therefore never get a confirmed ENE diagnosis. The AI provides 3D visual heat maps to guide treatment planning and helps doctors identify high-risk patients who may need more aggressive therapy, as well as low-risk patients who could receive less intensive treatment.