Deep Learning ECG as a Non-Invasive Window into Kidney Function and Electrolytes

Chin Lin Speaker Deep Learning Electrocardiography as a Non-Invasive Window into Kidney Function and Electrolyte DisturbancesRecent advances in deep learning have transformed the standard 12-lead electrocardiogram (ECG) from a tool for rhythm interpretation into a scalable physiologic sensor capable of detecting systemic disease. In this lecture, we will review the development and clinical translation of AI-enabled ECG models for dyskalemia detection and renal-function estimation. Using large real-world cohorts, convolutional and attention-based neural networks have demonstrated high accuracy for identifying moderate-to-severe hyperkalemia and hypokalemia directly from ECG waveforms, frequently preceding laboratory confirmation and outperforming clinician interpretation. Beyond electrolyte detection, AI-ECG signatures were also associated with adverse outcomes, cardiovascular risk, and future chronic kidney disease progression, even among patients with apparently normal laboratory findings. We will further discuss pragmatic deployment studies showing how real-time AI-ECG alerts integrated into emergency department workflows can accelerate treatment decisions for life-threatening hyperkalemia. Finally, the talk will address how signal-based AI can complement EHR-based prediction models in critical-care nephrology, including issues of interpretability, calibration drift, implementation, and multimodal foundation-model integration.