|| Checking for direct PDF access through Ovid
Purpose: Promising data have been obtained with non-ionizing, patient-friendly lung ultrasound (LUS) evaluation of B-lines (also called ultrasound lung comets), an emerging echographic sign of pulmonary congestion. Our aim was to assess whether a soft computing-based analysis of B-lines could objectively classify the severity of pulmonary congestion.Methods: Seventy-five pts with pulmonary congestion, as documented by clinical and instrumental data, and 40 control subjects underwent LUS examination. A soft computing-based modeling approach was applied to develop a model predictive of the severity of pulmonary congestion. The model was identified by means of a self-organizing Artificial Neural Network (ANN), a mathematical model, whose distributed adaptable parameters are modified through a learning process according to a real dataset. The dataset was composed of features extracted from B-lines, in terms of levels of severity of pulmonary congestion. In order to explore the prognostic values of the selected patterns and test the performance of the methodology, a cross-validation procedure was also applied.Results: The ANN based method was able to discriminate 4 levels of severity of pulmonary congestion (absence, mild, moderate, and severe). In particular, the model correctly identified 100% of pts with severe pulmonary congestion and 100% of pts without congestion. It also identified 92.8±1.5% of pts with mild pulmonary congestion, misclassifying 7.2%±0.8 as belonging to the moderate level, and identified 78.5±1% of pts with moderate pulmonary congestion, misclassifying 7±.2% as belonging to the mild level, and 14.5±.6% as belonging to the severe level.Conclusions: Soft computing analysis of B-lines allows objective categorization of the severity of pulmonary congestion.