This technology utilizes a sophisticated machine learning algorithm to detect intermediate- and high-risk pulmonary embolism (PE) using data from a standard 12-lead electrocardiogram (ECG). By enabling rapid identification of severe blood clots at the point of care, it facilitates earlier intervention and significantly reduces diagnostic delays for critically ill patients.
Description
The invention is a feature-based artificial intelligence model that analyzes raw waveform data from a non-invasive 12-lead ECG to identify patients at high risk for severe pulmonary embolism. Unlike standard diagnostic methods that may overlook subtle signs, this algorithm computes 419 distinct features from the ECG signal—including lead-specific measurements, beat-to-beat variability, and power spectral density, to generate a predictive risk score. It specifically employs a CatBoost supervised learning model, which utilizes an ensemble of decision trees to refine predictions based on the most significant clinical features.
Applications
- Emergency Department Triage: Integration into hospital triage software to prioritize chest pain and dyspnea patients for urgent diagnostic imaging.
- Pre-Hospital Diagnostics: Use by paramedics and emergency medical services (EMS) to identify severe PE patients before they reach the hospital.
- Diagnostic Software Licensing: Integration into existing 12-lead ECG hardware and software platforms as a value-added AI diagnostic feature.
- Clinical Decision Support Systems: Implementation within electronic health record (EHR) systems to provide automated risk alerts for severe cardiovascular events.
- Specialized PE Centers: A tool for Pulmonary Embolism Response Teams (PERT) to quickly assess patient severity and determine the need for advanced invasive treatments.
Advantages
- Rapid Triage: Enables immediate identification and expedited care for patients at the highest risk of severe pulmonary embolism.
- High Diagnostic Accuracy: Demonstrates robust performance with an Area Under the Curve (AUC) of 0.90 in identifying intermediate- and high-risk PE.
- Cost-Effective Screening: Offers a significantly less expensive alternative to immediate CT scanning for initial risk assessment.
- Seamless Integration: Can be easily applied in emergency departments or by paramedics as ECGs are already part of routine cardiopulmonary protocols.
- Improved Safety: Assists in reducing diagnostic delays and preventing mortality associated with severe, undetected blood clots.
Invention Readiness
A functional software prototype of the algorithm currently exists. The technology has been retrospectively validated using a large dataset from a major medical center, including 759 severe PE cases and over 20,000 controls, demonstrating a high degree of predictive accuracy (AUC 0.90). Future development plans include prospective clinical testing and evaluation of the model's performance across broader patient cohorts to refine its triage capabilities.
IP Status
Patent Pending
