Abstract
Knee Osteoarthritis (KOA) is a leading cause of disability globally. However, most prediction models focus on radiographic progression rather than symptomatic changes, despite structural worsening correlating only weakly with patient-reported pain. We present a multimodal mid-fusion Transformer that classifies nine-year KOA pain trajectories into four clinically motivated phenotypes (low-stable, improving, worsening, high-persistent) from tabular clinical data, X-ray, and Magnetic Resonance Imaging (MRI), using data from 3,751 participants over two annual visits in the Osteoarthritis Initiative (OAI). Patients with incomplete imaging or visits are retained without imputation via a hierarchical masking scheme that treats missingness as a learnable signal. Across 14 controlled-modality experiments, the best configuration achieves a weighted Area Under the Receiver Operating Characteristic Curve (AUC) of 0.89, comparable to binary-target state-of-the-art models despite the higher task complexity. Tabular data drives predictive performance across all configurations. Dynamic temporal input incorporating latent delta tokens improves worsening-class AUC from 0.61 to 0.76. These results indicate that pain trajectories are learnable from heterogeneous longitu- dinal data and support pain-centered prognostic modeling as a basis for earlier and targeted intervention in KOA.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MultiTab_020.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=cAoMll4dP6
BibTex
@InProceedings{GilAmn_Predicting_MICCAISAT2026,
author = { Gillani, Amna AND Schönecker, Jonathan AND Starck, Sophie AND Valle, Christina AND von Eisenhart-Rothe, Rüdiger AND Rueckert, Daniel AND Hinterwimmer, Florian},
title = { { Predicting Knee Osteoarthritis Pain Trajectories from Multimodal Longitudinal Data } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17263},
month = {pending},
page = {pending}
}
