Deep learning for automated detection of B-lines in lung ultrasonography

The application of ultrasound imaging to the diagnosis of lung diseases is gaining attention. Of particular interest are several imaging-artifacts, e.g., A and B line artifacts. A-lines are hyperechoic horizontal lines, which are substantially visualized across the entire image and parallel to pleural-line. They represent the normal pattern of the lung if pneumothorax is excluded. Differently, B-line artifacts correlate with pathology and are defined as hyperechoic vertical artifacts, which originate from a point along the pleura-line and lie perpendicular to the latter. Their presence has been linked to an increase in extravascular lung water, interstitial lung diseases, non-cardiogenic lung edema, interstitial pneumonia and lung contusion. In this work, we describe a method aimed to support the clinicians by automatically identifying the frames of an ultrasound video where B-lines are found. To this end, we employ modern deep learning strategies and train a fully convolutional neural network to perform this task on b-mode images of dedicated ultrasound phantoms (Demi et al., Sci. Rep. 2017). We moreover calculate neural attention maps that visualize which components in the image triggered the network, thereby offering simultaneous localization. Future work includes characterization of the detected B-lines to enable adequate phenotyping of various lung pathologies.