Optimizing the Diagnostic Algorithm for Pulmonary Embolism in Acute COPD Exacerbation Using Fuzzy Rough Sets and Support Vector Machine

Abstract

Aiming to optimize the diagnosis of pulmonary embolism (PE) in patients with acute exacerbation of chronic obstructive pulmonary disease (AECOPD), we conducted a retrospective study enrolling 185 AECOPD patients, of whom 90 were diagnosed with PE based on computed tomography pulmonary angiography (CTPA). Ten characteristic indicators and 27 blood indicators were extracted for each patient. First, we quantified the importance of each indicator for diagnosing PE in AECOPD using fuzzy rough sets (FRS) and selected the more important indicators to construct a support vector machine (SVM) diagnosis model called FRS-SVM. The performance of the proposed diagnosis model on the test sets was compared to that of the logistic regression model. The average accuracy and area under the curve (AUC) of the proposed model for the test sets in 10 independent trials were 94.67% and 0.944, respectively, compared to 80.41% and 0.809 for the logistic regression model. Thus, we validated the higher accuracy and stability of the FRS-SVM for PE diagnosis in patients with AECOPD. This model improved the prediction probability before CTPA and can be used in clinical practice to help doctors make decisions.

Keywords

• acute exacerbation of chronic obstructive pulmonary disease
• pulmonary embolism
• fuzzy rough sets
• support vector machine