PaCX-MAE: Physiology-Augmented Chest X-Ray Masked Autoencoder
Researchers have developed PaCX-MAE, a novel framework that integrates physiological data with chest X-ray (CXR) imaging for improved diagnostic models. This cross-modal distillation approach enhances CXR encoders by incorporating physiological priors, such as ECG and laboratory data, without requiring multimodal input during inference. Evaluations show PaCX-MAE significantly boosts performance across various benchmarks, particularly in tasks sensitive to physiological indicators, while also demonstrating strong label efficiency and anatomical fidelity. AI
IMPACT Enhances diagnostic accuracy by integrating multimodal data insights into unimodal AI models.