CardioScan AI: A Full-Stack Medical AI System for Predicting Left Ventricular Ejection Fraction Using 12-Lead ECG

CardioScan AI is a full-stack medical web application whose core function is to use artificial intelligence to predict Left Ventricular Ejection Fraction (LVEF) from 12-lead ECG scans. This project aims to address the limitations of traditional LVEF measurement, which relies on expensive imaging equipment (such as echocardiography) and requires professional operation, providing a low-cost and high-efficiency new approach for cardiac function assessment, and demonstrating the innovative application of deep learning in the field of medical diagnosis.

Left Ventricular Ejection Fraction (LVEF) is a core indicator for evaluating the pumping function of the heart, and it is crucial for heart failure diagnosis and treatment decisions. Traditional LVEF measurement relies on imaging examinations such as echocardiography, which has problems like high equipment cost, complex operation, and dependence on professional technicians. The emergence of CardioScan AI, which predicts LVEF by analyzing routine ECG data, brings new possibilities for cardiac function assessment.

CardioScan AI adopts a full-stack technical architecture, including three main layers:

1. Frontend Layer: Provides an intuitive interface for ECG upload, result display, and historical record management, focusing on user experience and data visualization in medical scenarios;
2. Backend Service Layer: Handles business logic, user authentication, and data management, using RESTful APIs to ensure data encryption and privacy security;
3. AI Inference Layer: Deploys deep learning models validated with clinical data to extract ECG signal features and predict LVEF.

Core technical innovations include:

• Signal Processing: Noise filtering, waveform segmentation, feature standardization;
• Model Design: 1D convolution (capturing local patterns) + LSTM/GRU (modeling time series) + attention mechanism (identifying key segments) + fully connected layer (outputting predicted values);
• Multi-task Learning: Simultaneously learning tasks such as arrhythmia detection and cardiac chamber size estimation to improve generalization ability.

The clinical value of CardioScan AI is reflected in:

1. Early Screening: Quickly identify high-risk groups with abnormal cardiac function in routine physical examinations to achieve early detection and intervention;
2. Resource-Constrained Areas: Use popular and low-cost ECG examinations to enable primary institutions to perform preliminary cardiac function assessments and narrow the medical gap;
3. Telemedicine: Integrate with wearable ECG devices to realize dynamic LVEF monitoring and support remote management of patients with chronic heart disease.

Challenges faced by the project and their solutions:

1. Data Quality Standardization: Differences in sampling rates and filtering settings of different devices are addressed by an adaptive preprocessing process to handle data in multiple formats, ensuring model robustness;
2. Model Interpretability: Integrate Grad-CAM visualization technology to highlight the regions in the ECG that contribute the most to the prediction, helping doctors understand the decision-making process;
3. Regulatory Compliance: Follow medical device development specifications and establish a sound data management and quality control system to lay the foundation for clinical translation.

As an open-source project, CardioScan AI provides the medical AI community with:

• Pretrained Models: Support transfer learning to accelerate the development of new models;
• Data Processing Tools: Standardized ECG signal processing process, which can be used in other ECG AI projects;
• Full-Stack Reference Implementation: Demonstrate how AI models can be integrated into a complete medical application.

The future prospects of CardioScan AI include:

1. Multi-Modal Fusion: Combine ECG, phonocardiogram, and clinical indicators to build a comprehensive cardiac function assessment system;
2. Personalized Modeling: Train dedicated models for specific groups such as children and athletes;
3. Real-Time Analysis: Optimize inference speed to support bedside instant detection;
4. Prognostic Prediction: Expand model capabilities to predict heart failure progression and cardiovascular event risks.

CardioScan AI demonstrates the great potential of artificial intelligence in the field of medical diagnosis. By combining deep learning with routine ECG, it provides a low-cost and high-efficiency new method for cardiac function assessment. As technology matures and clinical validation deepens, such AI tools are expected to become powerful assistants for doctors and ultimately benefit more patients.