Artificial intelligence-based diagnosis of acute pulmonary embolism: Development of a machine learning model using 12-lead electrocardiogram
See where this sits in the topic map →Summary AI-generated
- TL;DR
- This study develops and validates a machine learning model to diagnose acute pulmonary embolism (PE) using standard 12-lead electrocardiograms (ECGs).
- Problem
- Pulmonary embolism is a life-threatening condition where diagnostic uncertainty is high due to non-specific clinical presentations, typically requiring confirmation by computed tomography pulmonary angiography.
- Method
- Using 1,014 ECGs from emergency department patients suspected of having PE (911 for model development and 103 for validation), the researchers developed an ensemble neural network to predict the condition.
- Results
- The AI model achieved a specificity of 100% and a sensitivity of 50%, performing significantly better than standard clinical prediction rules (such as Wells, Geneva, PEGeD, and YEARS algorithms), which showed nearly no discriminative power in this comparison (AUC of 0.75).
- Contributions
- Not specified in the abstract.
- Limitations
- Not specified in the abstract.
- Takeaways
- An AI-based model using 12-lead ECGs can effectively aid in diagnosing acute pulmonary embolism in emergency settings, demonstrating exceptionally high specificity compared to traditional clinical scores.
- Applications
- The model can be applied in emergency departments to improve the specificity of acute pulmonary embolism detection over conventional clinical prediction rules.
- Topics
- Healthcare, Machine Learning, Medical Imaging and Diagnostics, Neural Networks
- For industry
- Healthcare and clinical decision support.
- Why it matters
- Advances AI-driven diagnostic tools for emergency medicine, potentially improving the accuracy of detecting life-threatening cardiovascular conditions.
Abstract
Pulmonary embolism (PE) is a life-threatening condition, in which diagnostic uncertainty remains high given the lack of specificity in clinical presentation. It requires confirmation by computed tomography pulmonary angiography (CTPA). Electrocardiography (ECG) signals can be detected by artificial intelligence (AI) with precision. The purpose of this study was to develop an AI model for predicting PE using a 12-lead ECG. We extracted 1014 ECGs from patients admitted to the emergency department who underwent CTPA due to suspected PE: 911 ECGs were used for development of the AI model and 103 ECGs for validation. An AI algorithm based on an ensemble neural network was developed. The performance of the AI model was compared against the guideline recommended clinical prediction rules for PE (Wells and Geneva scores combined with a standard D-dimer cut-off of 500 ng/mL and an age-adjusted cut-off, PEGeD and YEARS algorithm). The AI model achieves greater specificity to detect PE than the commonly used clinical prediction rules. The AI model shown a specificity of 100% (95% confidence interval (CI): 94–100) and a sensitivity of 50% (95% CI: 33–67). The AI model performed significantly better than the other models (area under the curve 0.75; 95% CI 0.66–0.82; p<0.001), which had nearly no discriminative power. The incidence of typical PE ECG features was similar in patients with and without PE. We developed and validated a deep learning-based AI model for PE diagnosis using a 12-lead ECG and it demonstrated high specificity. O tromboembolismo pulmonar é uma condição com elevada mortalidade, cujo diagnóstico constitui um desafio devido à frequente apresentação com sintomatologia inespecífica, requerendo confirmação diagnóstica por exame de imagem. Os sinais deletrocardiográficos podem ser detetados por mecanismos de inteligência artificial com precisão. O objetivo deste estudo foi desenvolver um modelo de inteligência artificial (IA) capaz de predizer embolia pulmonar (EP) com base no eletrocardiograma (ECG) de 12 derivações. Foram extraídos 1014 ECGs de doentes admitidos no serviço de urgência e em que foi realizada tomografia computorizada (TC) por suspeita de EP: 911 ECGs foram usados para o desenvolvimento do modelo e 103 ECGs foram usados para validação. Foi desenvolvido um modelo de IA com base no ECG de 12 derivações. O desempenho do modelo de IA foi comparado com os modelos de predição clínica de EP atualmente recomendados (score de Wells e de Geneva em combinação com um cut-off de D-dimer standard e com um cut-off ajustado à idade, algoritmo de PEGeD e algoritmo de YEARS). O modelo de IA demonstrou uma especificidade de 100% [95% CI: 94-100] e uma sensibilidade de 50% [95% CI: 33-67] para a deteção de EP. O modelo de IA demonstrou uma performance global significativamente superior aos restantes modelos (AUC 0,75; 95% CI 0,66-0,82; p<0,001), os quais demonstraram praticamente não ter poder discriminativo. A incidência dos achados eletrocardiográficos típicos de EP foi similar nos doentes com e sem EP. Neste estudo foi desenvolvido e validado um modelo de inteligência artificial para o diagnóstico de embolia pulmonar baseado no ECG de 12 derivações, que demonstrou elevada especificidade para o diagnóstico de embolia pulmonar em contexto de urgência.