Written by Chris Thom
Spoon Feed
Artificial Intelligence (AI) is poised for continuing growth in a variety of POCUS applications. This review article covers current barriers to AI use in POCUS.
Let’s learn just a bit about AI in POCUS
AI can seem complex, particularly in POCUS. This covers the basics and barriers to using it. AI models rely on a large quantity of human-expert labelled, annotated ultrasound imaging data for training and validation. This can be challenging for an AI network, given the variability in ultrasound imaging, including acquisition, anatomical variance, and artifacts. Ultrasound imaging often involves ill-defined anatomical borders, imaging of artifacts (e.g. – B-lines), and overlapping interfaces. This is more challenging than standard visual images that generally feature more defined structures and accompanying borders.
The present article discusses the potential of future “self-supervised” AI training, wherein manual human image labeling may not always be necessary. It also covers “explainability”, wherein the AI software allows users to comprehend the underlying analysis being performed, which adds clinician trust to the AI results or values being displayed.
How will this change my practice?
Perhaps the most useful applications of POCUS AI are the two examples discussed in the article, which are AI measurements of left ventricular ejection fraction and quantitative assessment of hip dysplasia. In the former, the AI model learns to recognize the endocardial border and tracks this through several cycles of systole and diastole. It then uses the common Simpson’s biplane method to calculate the EF based on the difference between volume dimensions in systole and diastole. See the accompanying image from the manuscript of this process. Also shown nearby is an example of the cNerve function of a GE ultrasound machine, where the femoral nerve is automatically recognized and shaded in yellow. These AI features required an immense amount of imaging data to train and validate. AI continues to find applications in POCUS, and this will empower clinicians by lowering the training bar for POCUS deployment in certain scenarios. It will also allow us to become more efficient as we learn to leverage these tools in our POCUS practice.
Source
Overcoming barriers in the use of artificial intelligence in point of care ultrasound. NPJ Digit Med. 2025 Apr 19;8(1):213. doi: 10.1038/s41746-025-01633-y. PMID: 40253547
