Introduction
Chest radiography, a widely accessible diagnostic tool, has traditionally been used to visualize anatomical structures in the chest. However, its potential in assessing cardiac function and valvular disease has remained largely unexplored. In a groundbreaking multi-institutional study, researchers developed and validated an artificial intelligence (AI)-based deep-learning model capable of simultaneously detecting cardiac functions and valvular diseases from chest radiographs.
Methodology
The study involved training, validating, and testing the deep-learning model using data collected from four different medical institutions. Chest radiographs and associated echocardiograms were obtained from three sites for internal testing and one site for external testing. The researchers assessed the model's performance by evaluating various metrics such as area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy.
Findings
A total of 22,551 radiographs and echocardiograms from 16,946 patients were included in the study. The external test dataset consisted of 3,311 radiographs from 2,617 patients, with an average age of 72 years. The model achieved impressive results in classifying different cardiac parameters. For example, it achieved an AUC of 0.92 for classifying left ventricular ejection fraction, 0.85 for tricuspid regurgitant velocity, 0.89 for mitral regurgitation, and so on.
Significance
This AI-based model demonstrates remarkable accuracy in classifying cardiac functions and valvular heart diseases using information extracted from digital chest radiographs. Moreover, it accomplishes this classification in a fraction of the time it takes to obtain similar information from echocardiography. Furthermore, the model has low system requirements and holds the potential to be continuously available in areas where specialized echocardiography expertise is limited.
Research Context
Prior to this study, no multi-institutional research had been conducted to estimate cardiac functions and valvular heart diseases from chest radiography. Existing studies were limited to individual valvular diseases and employed small, single-center datasets, which may lead to overfitting of the model.
Added Value
The study reveals that AI models can detect valuable information in radiological images that may be challenging for human observers to identify. This capability enables the model to efficiently classify information from chest radiographs, such as left ventricular ejection fraction and the presence of valvular heart disease. Notably, the model's low system requirements make it a feasible tool for regions where access to echocardiography specialists is limited.
Implications
The findings of this study suggest that chest radiographs, a static examination, can provide insights into cardiac function comparable to those obtained from dynamic examinations like echocardiography. By leveraging AI-based models, healthcare professionals can potentially expedite the diagnosis, monitoring, and treatment of cardiac function and valvular disease, leading to improved patient outcomes and cost-effectiveness. Continued advancements in this field hold promising prospects for transforming cardiac care worldwide.
Reference:
コメント