Lung ultrasound for point-of-care COVID-19 pneumonia stratification: computer-aided diagnostics in a smartphone. First experiences classifying semiology from public datasets
Aitor Almeida, Aritz Bilbao-Jayo, Lisa Ruby, Marga B. Rominger, Diego López–de–Ipiña, Jeremy Dahl, Ahmed ElKaffas, Sergio J. Sanabria
Abstract
Lung ultrasound (LUS) has demonstrated potential in managing pneumonia patients, and is actively used at the point-of-care in COVID-19 patient stratification. However, image interpretation is presently both time-consuming and operator-dependent. We explore computer-aided diagnostics of pneumonia semiology based on light-weight neural networks (MobileNets). For proof-of-concept, multi-task learning is performed from online available COVID-19 datasets, for which semiology (overall abnormality, B-lines, consolidations and pleural thickening) is annotated by two radiologists. Initial results suggest that individual indications can be classified with good performance in a smartphone. Neural networks may also help to reduce inter-reader variability and objectivize LUS interpretation, especially for early-stage pathological indications.