0
0This article introduces the Cardiac Sensing Foundation Model (CSFM), a versatile AI framework designed to standardize the analysis of diverse heart-related data. By training on a massive dataset from 1.7 million individuals, the model integrates different signal types like ECG and PPG alongside clinical text reports. Its transformer-based architecture allows it to adapt to various devices, ranging from medical-grade hospital monitors to consumer smartwatches. Extensive testing shows that CSFM outperforms traditional specialized models in tasks such as disease diagnosis, vital sign measurement, and clinical outcome prediction. Furthermore, the system acts as a powerful feature extractor and data reconstructor, capable of generating complex 12-lead heart signals from simple single-lead inputs. Ultimately, this foundation model provides a scalable and robust solution for improving cardiovascular care across both resource-rich and limited-resource environments.
References:
Gu X, Tang W, Han J, et al. Cardiac health assessment across scenarios and devices using a multimodal foundation model pretrained on data from 1.7 million individuals[J]. Nature Machine Intelligence, 2026, 8(2): 220-233.
