Automated Analysis of Ultrasound for the Diagnosis of Pneumothorax: A Systematic Review

Abstract

Automated analysis of lung ultrasound for pneumothorax detection is an emerging technology with potential applications across various clinical settings. This systematic review aims to evaluate the current evidence for the efficacy of automated analysis techniques in diagnosing pneumothorax using lung ultrasound images and videos. A literature search was conducted in the MEDLINE (Medical Literature Analysis and Retrieval System Online), Embase (Excerpta Medica Database), Web of Science, and Scopus databases up to July 5, 2024. Two reviewers screened articles, extracted data, and assessed the risk of bias using a modified QUADAS-2 (the revised version of the Quality Assessment of Diagnostic Accuracy Studies) tool. Ten studies met the inclusion criteria, with nine published after 2020. Eight studies were retrospective, and nine used physician analysis of ultrasound as the reference standard. Convolutional neural networks were the predominant classification method. The majority of studies analysed the presence or absence of lung sliding. Significant heterogeneity in study designs, outcome metrics, and methodology limited direct comparisons of efficacy. Automated analysis of lung ultrasound for pneumothorax detection shows promise; however, the field is still in its early stages. Challenges identified include the limitations of lung sliding as a diagnostic feature, variability in ultrasound hardware and protocols, and accounting for low-quality scans. Future research should focus on larger, more diverse datasets, standardized evaluation metrics, and robust external validation.