Abstract
Predicting patient-specific drug sensitivity from organoid cultures remains a critical bottleneck in precision oncology, typically requiring days of observation and costly endpoint assays, making the entire process very expensive. In this paper, we present a framework that relies on a world model for patient-derived organoids (PDO): it forecasts their morphological evolution by predicting future self-supervised DINOv2 features from early time-lapse microscopy frames. In particular, masked autoencoders are used as a learned transition model to predict the future representations of the organoid’s state, using information from previous dynamics. By operating in the representation space of a vision foundation model rather than in pixel space, the method learns the latent dynamics of the organoid and captures semantically meaningful morphologicalchanges. Leveraging a private collection of colorectal and pancreatic cancer organoids, we evaluate the fidelity of these forecasted features through two distinct downstream tasks: a functional ATP-based drug response estimation and a generative last-frame reconstruction task that maps the predicted features back into pixel space. Our results demonstrate that predicted future features enable drug response estimation before the endpoint is physically reached – effectively rolling the world model forward to a virtual endpoint – while the reconstruction task confirms the model’s ability to preserve structural and morphological integrity. By forecasting future organoid states, our framework opens a new avenue for early, non-destructive chemosensitivity screening, potentially accelerating clinical decision-making in precision medicine.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MWM_031.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=c2qoppjVXi
BibTex
@InProceedings{MenTho_An_MICCAISAT2026,
author = { Menard, Thomas AND Gontran, Emilie AND Mathieu, Jacques R. R. AND Cartry, Jérôme AND Cournède, Paul-Henry AND Jaulin, Fanny AND Christodoulidis, Stergios AND Le Chevalier, Veronique AND Vakalopoulou, Maria},
title = { { An Organoid World Model: Forecasting Self-Supervised Feature Dynamics for Early Chemosensitivity Prediction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17262},
month = {pending},
page = {pending}
}