Artificial Intelligence for Differentiating the Causes of Dyspnea

Abstract

<bold>Introduction:</bold> Dyspnea, a frequent symptom in emergency departments (ED) with multifactorial causes, requires accurate diagnosis. Artificial intelligence (AI) models are emerging as promising tools to support clinical decision-making. <bold>Methods:</bold> This retrospective study analyzed data from 787 dyspneic patients at Cantonal Hospital of Baselland in 2022, including clinical parameters, vital signs, and laboratory results. Machine Learning models (Decision Trees, Random Forest, and Decision Tree Boosting) were developed to differentiate respiratory and cardiac causes, with performance assessed using accuracy, sensitivity, and specificity. <bold>Results:</bold> The most common diagnoses were decompensated heart failure (28.4%), pneumonia (26.4%), and COVID-19 (17%). Decision Tree Boosting achieved the highest accuracy (89% for binary classification, 69.2% for quadruple classification, and 49.1% for single diagnosis), followed by Random Forest (88%, 64%, and 50%, respectively), and Standard Decision Trees (84.66%, 57.81%, and 36.82%). Key diagnostic parameters included C-reactive protein, B-type natriuretic peptide, and cough. Comorbidities had limited predictive value due to their non-specific nature. <bold>Conclusions:</bold> AI models can help differentiate cardiac and respiratory causes of dyspnea. However, low sensitivity for specific diagnoses highlights the need for larger datasets and advanced algorithms to improve reliability, requiring further validation before clinical integration.

Ähnliche Arbeiten