PaCX-MAE: Physiology-Augmented Chest X-Ray Masked Autoencoder
Yancheng Liu, Kenichi Maeda, Manan Pancholy · 2026
Read PaCX-MAE: Physiology-Augmented Chest X-Ray Masked Autoencoder by Yancheng Liu, Kenichi Maeda, Manan Pancholy (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-01537.txt.
Description
Liu, Maeda, and Pancholy teach chest X-ray models from body signals. Diagnosis often joins imaging with physiological measurements. Deployed models usually work on one data type alone. PaCX-MAE injects physiological priors into X-ray encoders. It stays strictly unimodal when used at inference time. Masked autoencoding in-domain forms the base training. A dual contrastive-predictive objective aligns X-ray features with paired signals. ECG and lab embeddings supply those paired signals.
Nine benchmarks show consistent gains over domain-specific MAE. Physiology-dependent tasks gain most from the method. Label efficiency is high in the 1 percent regime. Segmentation parity with MAE holds at the same time. Attention checks show focus on signs like the cardiac silhouette.
Catalog record
| CREATOR | Yancheng Liu, Kenichi Maeda, Manan Pancholy |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.01537