Graph Representation Learning of Longitudinal Medical Imaging Trajectories for Treatment Response Prediction

Jul 16, 2026

Upload by Violet

In breast cancer, pathological complete response (pCR) serves as a clinically meaningful surrogate for long-term outcomes, yet neoadjuvant chemotherapy response varies widely across patients, making individualized prediction difficult. A imaging-based 3D spatio-temporal framework is introduced that combines a state-of-the-art graph neural network with relational temporal interaction modeling across DCE-MRI timepoints. Three novel complementary self-supervised objectives learn treatment trajectory representations. Experiments on 585 ISPY-2 patients show strong gains over vision and self-supervised baselines and provide ablations, including effects of available timepoints and inter-scan time differences.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

Introduction

Problem motivation and clinical need

Limitations of existing pCR prediction approaches

Related longitudinal modeling work

arXiv :2607 .049 12v 1 [ cs .CV] 6 Jul 2026 Graph Representation Learning of Longitudinal Medical Imaging Trajectories for Treatment Response Prediction Johannes Kiechle 1 ,2 ,3 ,7 , Richard Osuala2 ,4 , Daniel M. Lang3 , Stefan M. Fischer 1 ,2 ,3 ,7 , Ivana Janíčková5 , Karim Lekadir4 , Julia A. Schnabel 1 ,3 ,6 ,7 ,†, and Jan C. Peeken2 ,† 1 Technical University of Munich, 2 TUM University Hospital Rechts der Isar, 3 Helmholtz Munich, 4 Universitat de Barcelona, 5 Medical University of Vienna, 6 King’s College London, 7 Munich Center for Machine Learning Abstract. In patients with breast cancer, pathological complete response (pCR) has been established as a clinically meaningful surrogate marker for long-term outcomes. While commonly treated with neoadjuvant chemotherapy (NACT), effective treatment decision-making remains challenging, as therapeutic response can vary substantially across patients, calling for predictive models capable of accurately estimating individualized treatment response. To address this, we propose an imaging-based 3D spatio-temporal framework for treatment response prediction that integrates a state-of-the-art graph neural network with relational modeling of temporal interactions across timepoints alongside three novel complementary self-supervised treatment trajectory representation learning objectives. Experiments across a cohort of 585 patients from the public ISPY-2 dataset demonstrate that our method substantially outperforms both vision and self-supervised learning baselines across several classification metrics. Alongside establishing a breast cancer pCR prediction benchmark, we include a principled ablation of our method and further introduce and empirically assess the impact of the available number of DCE-MRI timepoints per patient trajectory and the inclusion of inter-scan time-differences. Overall, our study substantiates the utility of clinically meaningful longitudinal medical imagaging modeling for predicting NACT-induced pCR. We will publicly share our code repository and a user-friendly PyPI library for dataset curation upon publication, effectively promoting reproducible open-source research. Keywords: GNNs · Longitudinal · Self-Supervised · Breast Cancer 1 Introduction With an estimated 2.3 million women diagnosed with breast cancer annually [1], optimizing treatment remains a clinical priority where neoadjuvant chemother-† Shared senior authorship. 2 Kiechle et al. Fig. 1. Method overview. Longitudinal image-derived latent features are represented as a directed acyclic graph and aggregated through a graph neural network (GNN) projection head to obtain a compact patient-level embedding. Representation learning is guided by an asymmetric response-aware objective that combines (a) population-level alignment,(b) patient-level representation decorrelation, and (c) temporal consistency losses. The overall objective encourages discriminative separation between respondersand non-responders while preserving the temporal structure of disease evolution. apy (NACT) crucially enables early response assessment to guide therapy adaptation and surgical planning [2] . Pathologic complete response (pCR), i.e. the absence of residual invasive disease in the breast and axillary lymph nodes at surgery following NACT [2], is commonly used to estimate NACT efficacy while being associated with improved long-term outcomes and event-free survival [3] . Reliable early prediction of pCR holds the promise of mitigating treatmentrelated toxicity and crucially supports informed clinical decision-making, including therapeutic escalation or de-escalation strategies. Moreover, early identification of non-responders may enable timely treatment modification and optimization of surgical intervention. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) plays an essential role in screening and assessment of response to NACT [4] . Accordingly, DCE-MRI–based methods for pCR prediction hav

FAQs

What clinical problem does the document address?

It targets reliable early prediction of pathological complete response (pCR) after neoadjuvant chemotherapy to support individualized treatment decisions.

How does the proposed method model longitudinal medical imaging?

It represents longitudinal, image-derived latent features as a directed acyclic graph and aggregates them with a graph neural network, using relational modeling of temporal interactions across timepoints.

What evidence is used to evaluate performance?

Experiments are conducted on a cohort of 585 patients from the public ISPY-2 dataset, showing improvements over vision and self-supervised learning baselines across multiple classification metrics.

Jul 16, 2026

Upload by Violet