Quantitative Lung Ultrasound Biomarkers for Automatic Stratification of Idiopathic Pulmonary Fibrosis: an in-vivo study

Idiopathic pulmonary fibrosis (IPF) is a progressive lung pathology that can occur in two distinct stages: chronic IPF and acute IPF exacerbation (IPFE). Although their histological conditions differ, their stratification through gold standard lung imaging remains challenging. Quantitative lung ultrasound (Q-LUS) has recently emerged as a promising, safe, and non-invasive imaging modality for characterizing the lung condition. Previous studies analyzed the frequency dependence of imaging artifacts features from different pathologies to enhance diagnostic specificity. However, these studies marginally explored IPF stratification, analyzing the mean values of specific spectral biomarkers computed over the lung ultrasound (LUS) scans of a patient. In this study, LUS images from a cohort of 32 patients with IPF are segmented and quantified into 12 spectral biomarkers with a fully automatic segmentation algorithm. Rather than compressing each spectral biomarker to its mean value, a deeper analysis of their distribution is performed. Five statistical parameters (minimum, 25th, 50th, and 75th percentile, and maximum) are derived for each spectral biomarker, yielding to 60 parameters per patient. Parameters are ranked by statistical significance to train nested k-fold Naïve Bayes classifiers. This method is compared with state-of-art approach, demonstrating accuracies improvement up to 13% in IPF stratification.