MorphologyFM: A Foundation Model for Morphology-Aware Representation Learning from ECG and Pulse Oximetry Waveforms

MorphologyFM is a multimodal foundation model pretrained on paired ECG and pulse oximetry waveforms from the MIMIC critical care database. Its self-supervised objective combines morphology-guided masking, cross-modal representation learning, and contrastive latent alignment. The paper evaluates the learned representations on arrhythmia classification, hypoxemia prediction, mortality prediction, and length-of-stay estimation, reporting consistent improvements over representative methods including Masked Autoencoders, contrastive learning, Barlow Twins, and JEPA. Joint ECG-SpO2 pretraining also outperforms single-modality pretraining, suggesting that waveform morphology and cross-modal physiological structure can improve transfer across continuous-monitoring tasks.