Reproduce “Curia-MAE: Multi-Modal Multi-Anatomy MAE Pre-Training for 3D Medical Image Segmentation” 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

Théo Danielou, Antoine Saporta, Léo Alberge, Corentin Dancette

Curia-MAE trains a reusable encoder for 3D medical segmentation by reconstructing masked scans and aligning features from a crop with corresponding features from the full scan. Trained on about 300,000 CT and MRI volumes, its frozen encoder raises lesion-task Dice over an MAE baseline, from 63.0 to 64.7, with a task-specific decoder trained on labeled data. The gain does not occur on every dataset or replace nnU-Net.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.