Automated Segmentation of M-Mode Lung Ultrasound Images Obtained from a Single-Element Wearable Ultrasonic Sensor

Lung ultrasound (LUS) is an increasingly popular imaging method for lung disease diagnosis. Adoption is limited by the availability of trained experts to interpret the artefact-based LUS images. With the recent advent of automated LUS interpretation methods using machine learning and wearable ultrasound technology, the benefits of LUS can potentially be leveraged in novel ways that overcome this training shortage. In this work, automatic segmentation methods are investigated for motion-mode LUS images of a lung phantom and eight healthy human subjects obtained using a single-element wearable ultrasonic sensor (WUS). The proposed DBSCAN cluster boundary estimation method resulted in 95.6% and 92.0% median Jaccard index for the phantom and in-vivo images, respectively. The WUS, in combination with the proposed segmentation methods, has potential to enable operator-independent, continuous, and long-term lung monitoring.