Self-Supervised Pretraining Improves Performance and Inference Efficiency in Multiple Lung Ultrasound Interpretation Tasks

TL;DR

This study demonstrates that self-supervised pretraining enhances lung ultrasound classification performance and inference efficiency across multiple tasks, especially with limited labeled data.

Contribution

The paper introduces a self-supervised pretraining approach that improves multi-task lung ultrasound classification and reduces inference time.

Findings

Pretrained models improved AUC by up to 0.061 on external test sets.

Inference time was reduced by 49% using pretrained features.

Pretrained models outperformed fully supervised models with only 1% labeled data.

Abstract

In this study, we investigated whether self-supervised pretraining could produce a neural network feature extractor applicable to multiple classification tasks in B-mode lung ultrasound analysis. When fine-tuning on three lung ultrasound tasks, pretrained models resulted in an improvement of the average across-task area under the receiver operating curve (AUC) by 0.032 and 0.061 on local and external test sets respectively. Compact nonlinear classifiers trained on features outputted by a single pretrained model did not improve performance across all tasks; however, they did reduce inference time by 49% compared to serial execution of separate fine-tuned models. When training using 1% of the available labels, pretrained models consistently outperformed fully supervised models, with a maximum observed test AUC increase of 0.396 for the task of view classification. Overall, the results indicate that self-supervised pretraining is useful for producing initial weights for lung ultrasound classifiers.