Combining quantitative and qualitative analysis for scoring pleural line in lung ultrasound

With the advancement of lung disease research and the wide application of lung ultrasound (LUS), it is essential to analyze various indicators in LUS images independently to aid in clinical diagnosis. In this paper, we proposed a quantitative and qualitative method for extracting, analyzing, and scoring pleural lines with different lesions in LUS images. The extraction module consists of customized cascaded detection and segmentation models based on convolution and multilayer perceptron (MLP). The analysis module uses eight textural and three morphological parameters to quantitatively analyze the features of two different output images from the localization and segmentation models, respectively. To qualitatively evaluate pleural lines with different severities in LUS images, the scoring module adopts four supervised machine learning classifiers, including support vector machine, k-nearest neighbor, random forest, and decision tree. We performed experiments on the 5390 LUS images acquired from Coronavirus Disease 2019 pneumonia patients using convex ultrasound probes. The experimental results demonstrated that our proposed line extraction method accurately detected and segmented pleural lines. The support vector machine classifier, which combined textural and morphological features as input, achieved optimal scoring performance with accuracy, sensitivity, specificity, F1 score, and AUC being 94.47%, 97.31%, 94.50%, 0.9457, and 0.9822, respectively. Compared with other models, our proposed method also proved to be more effective. Thus, our proposed method has great potential for clinical application in the analysis of LUS images and can aid in the diagnosis and treatment of lung disease.