Identification of B-lines in vivo lung ultrasound by the evaluation of characteristic parameters using raw RF data

Special artifact B-line of lung ultrasound is closely related to lung lesions and has become an important indicator to assist doctors to diagnose lung diseases. However, the identification of B-lines mainly depends on the experience and subjective judgment of clinicians. In order to assist physicians in making the correct diagnosis, a method based on RF raw data was proposed to identify B-line precisely and automatically. The t-test, Bayes classifier and decision tree classifier were used to test distinction and Identification correlation of several characteristic parameters including information entropy, permutation entropy, energy and nakagami between B-Line and non-B-Line regions. The significant difference and classification results of these characteristic parameters were compared. The results showed that the values of five characteristic parameters selected in this study have significant differences (P< 0.05) between B-Line regions and other regions. the identification effect of information entropy is the best of all characteristic parameters with Bayes classifier (accuracy =91.7%, AUC=0.90) and decision tree classifier (accuracy =91.6%, AUC=0.90). The results showed that the identification method of B-line based on RF data had important value in assisting clinical diagnosis, and the information entropy, mode entropy and permutation entropy could identify the B-line with high precision.