The artificial intelligence (AI) pipeline developed at a Montreal-based cancer center shows promise in detecting standard CT scans (ENEs) from standard CT scans of HPV-positive oropharyngeal cancer patients. This is a factor related to long-term outcomes.
The researchers tested the tool in 397 patients with clinical node-positive HPV-driven pharyngeal squamous cell carcinoma (OPSCC) who received chemoradiation therapy between 2009 and 2020. This feature, known as an imaging-based ENE, or IENE, is currently not part of the official staging criteria, but is linked to distant failure and low survival risk.
Detection without pathologists
This tool identified iENEs with matching or exceeding accuracy than experienced neuroradiologists. That prediction was associated with overall survival, no recurrence survival, and remote control, and remained independently important after adjusting for established clinical factors. Patients with IENE predicted by the AI model had lower 3-year survival rates (83.8%) compared to patients (96.8%).
The 8th edition of the American Joint Committee on Cancer (AJCC) staging manual excludes iene from staging HPV-positive cancers, but it is expected to include it in the upcoming 9th edition. However, detection of ENEs remains highly subjective and is often not available outside of tertiary care settings. This prompted interest in AI as a reproducible alternative that could support treatment decisions at the centre without neuroradiology expertise.
A two-stage pipeline for clinical prediction
The researchers developed a two-part system. First, there is the 3D neural network segmented gross tumor volume (GTV) of pathological lymph nodes. Second, the machine learning classifier used radioactive features to identify the presence or absence of iene. The segmentation model achieves an average dice similarity coefficient of 0.74, showing good overlap with expert contours. In IENE prediction, the optimal performance model used Lasso feature selection and xgboost classification to reach an AUC of 0.81.
To assess clinical impact, the team compared oncological outcomes between patients with or without AI predictive IENE. Multivariable analysis showed that AI-IENE was a significant predictor of worse overall survival (adjusted hazard ratio of 2.82), survival without recurrence (AHR 4.20), and remote control (AHR 12.33). In both cases, the model outperformed the radiologist's ratings when tested against survival data.
Clinical relevance and future applications
This model was trained with radiologist-determined IENE labels, but showed a stronger correlation with survival outcomes than the original human assessment. This suggests that the tool may identify imaging features linked to node aggression that is less obvious to the human eye.
The authors suggest that AI-driven IENE classifications can help clinicians stratify patients for treatment intensity. For example, those with predicted IENE may not be suitable for treatment deablative protocols, but IENE-negative patients may consider the intensity reduction option. Importantly, the study authors emphasize that further external validation is required and that the tool is not intended to change clinical pathways until more data is available.
Limitations and next steps
This single-center retrospective study may use radiologist IENE assessments as reference standards rather than pathological confirmation, potentially limiting generalizability. This model evaluated only the largest nodes per patient. This may miss a small lesion in ENE. Additionally, the dataset lacks demographic diversity and requires external validation of multicenter cohorts prior to clinical integration.
Despite these constraints, researchers suggest that integrating AI tools into clinical workflows can provide consistent, accessible image analysis to improve decision-making and reduce diagnostic variability. Future research will focus on validation, performance of various imaging systems as a whole, and incorporation into clinical trials.
reference: Dayan GS, HéniqueG, Bahig H, et al., Artificial Intelligent Models for Imaging-Based Extrinsic Detection and Result Prediction in Human Papilloma Virus-Positive Oropharyngeal Cancer. Jama Otolaryngol Head Neck Surg. , 2025. doi:10.1001/jamaoto.2025.3225
This content contains text generated with AI support. You can find the AI policy for your technology network here.
