Learning Generic Lung Ultrasound Biomarkers for Decoupling Feature Extraction from Downstream Tasks

Abstract

Contemporary artificial neural networks (ANN) are trained end-to-end, jointly learning both features and classifiers for the task of interest. Though enormously effective, this paradigm imposes significant costs in assembling annotated task-specific datasets and training large-scale networks. We propose to decouple feature learning from downstream lung ultrasound tasks by introducing an auxiliary pre-task of visual biomarker classification. We demonstrate that one can learn an informative, concise, and interpretable feature space from ultrasound videos by training models for predicting biomarker labels. Notably, biomarker feature extractors can be trained from data annotated with weak videoscale supervision. These features can be used by a variety of downstream Expert models targeted for diverse clinical tasks (Diagnosis, lung severity, S/F ratio). Crucially, task-specific expert models are comparable in accuracy to end-to-end models directly trained for such target tasks, while being significantly lower cost to train.

Artificial Intelligence Neural Network Consistently Interprets Lung Ultrasound Artifacts in Hospitalized Patients: A Prospective Observational Study

“…A previously published artificial intelligence neural network which has been used to analyze lung ultrasound artifacts [12,13] by employing a Temporal Shift Module (TSM) [14] was trained using 485 ultrasound clips from 142 research subjects. The previously published model characterized A and B line artifacts with an accuracy of 76.4% and a precision of 70.8% [13]. The TSM model is video-based model that jointly analyzes a group of frames belonging to a video clip in order to simultaneously predict A-lines and B-lines [14].…”