AMO-ENE: Attention-based Multi-Omics Fusion Model for Outcome Prediction in Extra Nodal Extension and HPV-associated Oropharyngeal Cancer

A multi-institutional research team has developed AMO-ENE, a novel AI pipeline designed to automate the prediction of treatment outcomes for HPV-positive oropharyngeal cancer (OPC). The system addresses a critical clinical gap: the current omission of extranodal extension (ENE)—a key prognostic factor—from formal staging criteria due to difficulties in manual detection on CT scans. AMO-ENE uses a hierarchical 3D semi-supervised segmentation model to automatically detect and delineate ENE from standard radiotherapy planning CT scans, overcoming issues like low contrast and inconsistent manual annotations.

From these automated segmentations, the model extracts a comprehensive set of radiomic and deep-learning features to train a classifier for imaging-detected ENE. This predicted ENE status is then fused with primary tumor characteristics and clinical data within a multimodal, attention-based architecture to forecast patient outcomes. The model was rigorously validated on an internal cohort of 397 HPV-positive OPC patients treated between 2009 and 2020.

The results are striking for clinical translation. For 2-year outcome prediction, AMO-ENE achieved an area under the curve (AUC) of 88.2% for metastatic recurrence, 79.2% for overall survival, and 78.1% for disease-free survival, significantly surpassing existing baseline models. The corresponding concordance indices were 83.3%, 71.3%, and 70.0%, respectively. This performance demonstrates the model's feasibility as a tool for risk stratification and personalized treatment planning, moving AI from pure detection into the realm of prognostic decision support.