Reproduce “Spatio-Temporal Analysis of Patient-Derived Organoid Videos Using Deep Learning for the Prediction of Drug Efficacy” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Leo Fillioux, Emilie Gontran, Jérôme Cartry, Jacques RR Mathieu, Sabrina Bedja, Alice Boilève, Paul-Henry Cournède, Fanny Jaulin and 2 more
Researchers developed a deep learning framework for spatio-temporal analysis of patient-derived organoid videos to predict drug efficacy, specifically the ATP readout. This non-destructive method integrates foundation models for automated image analysis, achieving a Mean Absolute Percentage Error of 0.1755 and a Pearson correlation of 0.9214 for ATP prediction.
Key insights
- Self-supervised vision transformer features (DINOv2) provide superior representations for organoid characteristics relevant to drug efficacy prediction compared to classical radiomics or ImageNet-pretrained features.
- A learnable temporal attention mechanism is crucial for dynamically weighing different time points in the video sequence, with both early and late frames identified as most informative for predicting drug response.
- Leveraging foundation models like SAM for segmentation, coupled with temporal propagation, effectively automates complex biological image analysis, reducing the need for extensive manual annotations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.