Quantitative Lung Ultrasound Spectroscopy Classification Performance in Differentiating CPE, Pneumonia, and PF, a Comparative Classifiers’ Analysis

Lung ultrasound (LUS) is widely adopted to assess the state of lung surface, which is mainly performed by analyzing imaging artifacts. Of particular interest is the dependence of vertical artifacts (VAs) on frequency, which was demonstrated to carry important diagnostic information able to improve LUS specificity. In this multicenter study, raw radiofrequency (RF) data were acquired from 114 patients. Specifically, an Ultrasound Advanced Open Platform (ULA-OP) was used to collect RF data, and a multifrequency approach implemented with linear and convex probes. VAs were segmented, and their frequency content analyzed. The potential of three features in discriminating cardiogenic pulmonary edema (CPE), pneumonia, and pulmonary fibrosis (PF) was assessed. Indeed, these features were fed to different classifiers, both binary and multiclass. Results show how the use of mean I<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TOT</inf> leads to average accuracy up to 82.4% and 60.3% for binary and multiclass classifiers, respectively.