AMO-ENE: Attention-based Multi-Omics fusion model for outcome prediction in Extra Nodal Extension and HPV-associated oropharyngeal cancer
Gautier Hénique, William Le, Gabriel Dayan MD, Coralie Brodeur, Kristoff Nelson MD, Apostolos Christopoulos MSc, MD, Edith Filion MD, Phuc-Felix Nguyen-Tan MD, Laurent Letourneau-Guillon MD, MSc, Houda Bahig MD, PhD, Samuel Kadoury PhD · 2026
Read AMO-ENE: Attention-based Multi-Omics fusion model for outcome prediction in Extra Nodal Extension and HPV-associated oropharyngeal cancer by Gautier Hénique, William Le, Gabriel Dayan MD, Coralie Brodeur, Kristoff Nelson MD, Apostolos Christopoulos MSc, MD, Edith Filion MD, Phuc-Felix Nguyen-Tan MD, Laurent Letourneau-Guillon MD, MSc, Houda Bahig MD, PhD, Samuel Kadoury PhD (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-09280.txt.
Introduction
Hénique and colleagues build AMO-ENE, a model that blends many data types to predict outcomes in HPV-linked throat cancer with extranodal extension. That spread beyond nodes is a new warning sign left out of staging, hindered by uneven outlining, low node contrast on CT, and slow hand labeling. Their auto pipeline outlines nodal extension from treatment CT scans, extracts image and deep features, and grades seen extension. Combined with main tumor features, it predicts outcomes in 397 patients, reaching 88 percent AUC for metastatic recurrence at two years.
Catalog record
| CREATOR | Gautier Hénique, William Le, Gabriel Dayan MD, Coralie Brodeur, Kristoff Nelson MD, Apostolos Christopoulos MSc, MD, Edith Filion MD, Phuc-Felix Nguyen-Tan MD, Laurent Letourneau-Guillon MD, MSc, Houda Bahig MD, PhD, Samuel Kadoury PhD |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 73531 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.09280
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
AMO-ENE: Attention-based Multi-Omics fusion model for outcome prediction in Extra Nodal Extension and HPV-associated oropharyngeal cancer Gautier Hénique, William Le, Gabriel Dayan MD, Coralie Brodeur, Kristoff Nelson MD, Apostolos Christopoulos MSc, MD, Edith Filion MD, Phuc-Felix Nguyen-Tan MD, Laurent Letourneau-Guillon MD, MSc, Houda Bahig MD, PhD, Samuel Kadoury PhD Abstract Extranodal extension (ENE) is an emerging prognostic factor in human papillomavirus (HPV)-associated oropharyngeal cancer (OPC), although it is currently omitted as a clinical staging criteria. Recent works have advocated for the inclusion of iENE as a prognostic marker in HPV-positive OPC staging. However, several practical limitations continue to hinder its clinical integration, including inconsistencies in segmentation, low contrast in the periphery of metastatic lymph nodes on CT imaging, and laborious manual annotations. To address these limitations, we propose a fully automated end-to-end pipeline that uses computed tomography (CT) images with clinical data to assess the status of nodal ENE and predict treatment outcomes. Our approach includes a hierarchical 3D semi-supervised segmentation model designed to detect and delineate relevant iENE from radiotherapy planning CT scans. From these segmentations, a set of radiomics and deep features are extracted to train an imaging-detected ENE grading classifier. The predicted ENE status is then evaluated for its prognostic value and compared with existing staging criteria. Furthermore, we integrate these nodal features with primary tumor characteristics in a multimodal, attention-based outcome prediction model, providing a dynamic framework for outcome prediction. Our method is validated in an internal cohort of 397 HPV-positive OPC patients treated with radiation therapy or chemoradiotherapy between 2009 and 2020. The segmentation model achieved a mean Dice score of 78.4 ( ± \pm 7.5), which improved to 83.5 83.5 ( ± 4.1 \pm 4.1 ) following a component selection method. The ENE classification model reached 81.6 % 81.6\% ( ± \pm 5.7) in AUC for general iENE detection, and 89.9 % \% ( ± \pm 5.0) when identifying high-grade (grade 3) ENE. For outcome prediction at the 2-year mark, our pipeline surpassed baseline models with 88.2 % \% ( ± \pm 4.8) in AUC for metastatic recurrence, 79.2 % \% ( ± \pm 7.4) for overall survival, and 78.1 % \% ( ± \pm 8.6) for disease-free survival. We also obtain a concordance index of 83.3 % \% ( ± \pm 6.5) for metastatic recurrence, 71.3 % \% ( ± \pm 8.9) for overall survival, and 70.0 % \% ( ± \pm 8.1) for disease-free survival, making it feasible for clinical decision making. 1 Introduction Extranodal extension (ENE) in oropharyngeal carcinoma (OPC) refers to the spread of malignant cells beyond the capsule of metastatic lymph nodes [ 2 ] , and is an indicator of tumor aggressiveness, as well as of a greater risk for invasion into adjacent tissues. Traditionally, ENE is confirmed histopathologically (pENE) through detailed examination following lymph node dissection, a time-intensive process. Presence of ENE is associated with escalated therapeutic regimens, including intensification of adjuvant and chemo-radiation protocols, as well as closer surveillance protocols [ 17 ] , which, while aiming to curb disease progression, also increases the burden of treatment-related toxicity on patients. However, a significant proportion of patients with OPC are currently treated with upfront chemoradiation without undergoing neck dissection, resulting in the absence of formal pathological lymph node evaluation, thus motivating the use of imaging as a surrogate marker of ENE. To expedite the evaluation process, imaging-based gradation systems for imaging detectable ENE (iENE) on head and neck CT scans have been developed in recent years, such as the five grade scale proposed by [ 17 , 19 , 18 ] . However, these lack standardization, and their clinical utility is hindered by inter-observer variability and inconsistent segmentation boundaries. Figure 1 illustrates the four-tier iENE classification framework: grade 0— no radiologic evidence of ENE; grade 1— minor extension (through the capsule into surrounding fat); grade 2— coalescing matted nodal mass with disappearance of inter-nodal planes; and grade 3— gross invasion into surrounding anatomical structures (adapted from [ 19 , 18 ] ). Therefore, despite increasing evidence advocating for the incorporation of ENE status into the staging of HPV-associated OPC, numerous challenges ranging from diagnostic inconsistency to a lack of universal criteria continue to impede its widespread and reliable integration into oncologic practice. Figure 1 : Illustration for iENE status grading system. Nodes are presented with white borders, red healthy tissue background and black tumoral content. Grade 0 indicates metastatic pathological node with no signs of extension. Grade 1 indicates extension through the nodal capsule into surrounding fat. Grade 2 characterizes coalescent nodal masses with loss of internodal planes. Grade 3 is defined by overt invasion of surrounding structures and tissues. In case of multiple ENE, the maximum grade is reported. In recent years, there has been growing interest in leveraging machine learning (ML) techniques to extract valuable prognostic insights from medical imaging, providing a way to extract predictive information beyond subjective image interpretation and potentially enhancing clinical guidelines with more precise and reliable image analysis. By leveraging the complementary predictive value of multiple modalities and their interactions, patient-specific treatment strategies could be integrated and result in reduced toxicities ( [ 4 ] ). In this paper, we propose AMO-ENE, a fully automated pipeline using an attention-based multi-organ fusion model for for outcome prediction in head and neck oropharyngeal cancer, focusing on primary tumors and their association with metastatic extranodal extension (ENE) in CT images. This study proposes a novel prognostic pipeline that includes an ENE segmentation model and introduces a multimodal and multi-omic attention fusion model for multi-bin outcome risk prediction. Our objective is to propose an end-to-end, clinically applicable pipeline, with efforts on exploring stronger deep modality fusion algorithms paired with deep survival methods. This paper’s contributions include the following: • Automation of pathological node segmentation on radiation planning CTs, comparing state-of-the-art methods, and subsequent iENE grade classification based on complementary feature sets; • Introduction of an attention-based multi-omics fusion model for 2-year risk assessment in HPV-associated OPC treated with radiation therapy; • Adaptation of the proposed model to multi-bin risk modeling for a deep learning based-longitudinal risk estimation of outcomes in HPV-associated OPC; • Evaluation of extra nodal extension prognostic value using statistical survival analysis on a cohort of 397 OPC patients. 2 Related Works 2.1 Lesion segmentation Medical image segmentation is a popular task in oncological research, as robust tumor detection and delineation may enable automated screening and treatment planning processes. In recent years, several approaches were proposed to automatically retrieve organs at risk (OARs) from planning images [ 33 ] , however they often exclude nodal structures in this category. In addition, there are no methods to our knowledge for automated extra nodal extension segmentation in the literature, which may be explained by the lack of iENE status integration in clinical guidelines for HPV-associated OPC and by the labour intensive annotation task it demands. Since its introduction in 2012 [ 25 ] , deep convolutional neural networks (CNN) have established themselves as a standard for computer vision tasks. Adapted to the popular encoder-decoder "U" style architecture [ 35 ] , it varies to different field of views by harnessing the strong spatial awareness of convolution masks in an encoder-decoder fashion for 3D segmentation