ECGomics: An Open Platform for AI-ECG Digital Biomarker Discovery

Figures

• 

  Fig. 1.. Taxonomic parallel between Genomics and ECGomics. Genomics organizes DNA information into Structural, Mutational, Functional, and Comparative domains. Correspondingly, ECGomics deconstructs cardiac signals into Structural, Intensity, Functional, and Comparative dimensions, enabling transformation of raw electrocardiography (ECG) signals into multidimensional digital biomarkers.

• 

  Fig. 2.. ECGomics-driven workflow from digital biomarkers to clinical translation. The framework comprises 3 layers: (Left) ECGomics data layer for high-throughput extraction of structural, intensity, functional, and comparative ECGomics from raw electrocardiography (ECG) signals; (middle) biomedical analysis layer integrating ECG signatures with clinical, imaging, and multiomics data for deep phenotyping and multiomics association studies; (right) application layer translating insights into precision intervention, personalized therapy, and disease trajectory management.

• 

  Fig. 3.. Artificial intelligence-electrocardiography (AI-ECG) digital biomarker overview. This illustrates the classification logic from the underlying raw features to the high-level predictive model, mainly including: engineered biomarkers (corresponding to structural, intensity, and functional dimensions, such as waveform duration, spectrum, and heart rate variability), predictive biomarkers (corresponding to comparison dimensions, such as cardiac age and disease risk prediction), and deep biomarkers based on deep learning (such as various hidden layer embedded features).

• 

  Fig. 4.. User interface and workflow of the ECGomics platform. (A) Signal acquisition configuration (SampleRate, adcGain, and adcZero). (B) Data selection via sample library or custom .npy upload. (C) Raw electrocardiography (ECG) numerical display. (D) Twelve-lead ECG visualization with export function. (E) “Generate ECGomics” execution control. (F) Language switching (Chinese/English). (G) Output panel presenting results across 4 ECGomics dimensions, with CSV export option.

• 

  Fig. 5.. Workflow for portable electrocardiography (ECG) collection and ECGomics analysis. (A) Initiate data acquisition via the mobile application. (B) Record ECG signals using the handheld device with real-time waveform feedback. (C) Automated analysis generates a comprehensive report including waveform visualization, rhythm parameters (e.g., heart rate, PR interval, and QRS duration), and predictive metrics such as cardiac age and risk assessment.

• 

  Fig. 6.. Comparison of different electrocardiography (ECG) prediction workflows. This figure illustrates the methodological evolution and performance trade-offs across 3 ECG analysis paradigms. (Top) The Expert feature-driven workflow utilizes handcrafted features based on clinical knowledge, offering high interpretability but limited accuracy due to the omission of latent signal patterns. (Middle) The deep learning (DL)-driven workflow achieves superior accuracy through end-to-end feature learning but functions as a “black box” with low interpretability and high data dependency. (Bottom) The proposed ECGomics-driven workflow synergizes expert-defined morphology with deep latent representations. As shown in the bar charts (right), the ECGomics approach achieves a high balance between accuracy and interpretability while maintaining a low data requirement, demonstrating its superior utility for precision medicine applications.