Identification of alveolar recruitment patterns in lung ultrasound videos

PASSONI, LUCÍA ISABEL; MESCHINO, GUSTAVO JAVIER; ACOSTA, CECILIA; TUSMAN, GERARDO

Piriápolis

Congreso; XXII Congreso de Bioingeniería; 2020

Sociedad Argentina de Bioingeniería (SABI)

Ultrasound has emerged as relevant technique in the field of anesthesia and critical care. Pulmonary ultrasound (UP) is a dynamic, non-invasive diagnostic tool, suitable for patients under mechanical ventilation. General anesthesia is associated with total or partial collapse of the lung (atelectasis) in a high percentage of patients. Recruitment maneuvers (MR) are dynamic and physiological processes of re-aeration of a gas-free lung region, with positive pressure ventilation. To perform the evaluation with UP images of the result of a recruitment, ultrasound patterns are identified that allow the diagnosis of different pleuro-pulmonary pathologies. This work advances in the direction of developing a monitoring support tool during the MRs, assisting in the interpretation of the dynamics of ultrasound patterns that are recorded while they are performed. This work proposes the processing of UP video sequences, feeding an artificial neural network with deep learning architecture, retrained by transfer learning. UP videos acquired in different patients to whom recruitment maneuvers were applied were processed, resulting in videos corresponding to different pulmonary situations. An ultrasound imaging specialist labeled some video frames to get training information for the system. For each frame, membership probabilities values for each class are obtained. A summary analysis of each video is also obtained. Despite the complexity of ultrasound images, the network proposed can detect different states with great efficiency and effectiveness. The system constitutes a contribution in the processing of pulmonary ultrasound videos, adding computational information to the visual and quantitative analysis that have been proposed so far.