Informatics at the Bedside: Intelligent Systems in Point-of-Care Ultrasound

…Like humans, they improve as they encounter more data. Unlike humans, once trained, these systems can process enormous amounts of information and provide answers in a fraction of the time…

Introduction and Aims

Point-of-Care Ultrasound (POCUS) is a diagnostic imaging technique often used at the bedside in a variety of settings including emergency care, cardiology, intensive care, anesthesiology and obstetrics among others. Alongside the rapid evolution of machine learning (ML) and deep learning (DL) algorithms in the health informatics space, “intelligent” POCUS systems are concurrently transforming. From guiding novice trainees to automating clinical calculations, these algorithms are making ultrasound faster, more accurate, and more accessible. This post aims to provide a window into the innovations at the intersection of informatics and POCUS driving that transformation as well as what this integration means for clinicians and providers going forward.

Brief Conceptual Review

Clinical Informatics refers to the use of data, technology, and algorithms to enhance decision-making, diagnostics, and care delivery. Machine learning and deep learning, subfields of clinical informatics, use algorithms that learn patterns from data rather than following fixed instructions.

Intelligent POCUS Systems

Intelligent ultrasound systems are best understood as informatics tools - technologies that combine clinical data with algorithmic processing to guide users and improve care at the bedside. Intelligent POCUS systems are revolutionizing ultrasound training by embedding ML and DL algorithms that provide real-time, actionable feedback to novice users. Ultrasound training requires developing the visuospatial awareness to link a two dimensional grayscale cross-section into a patient’s three dimensional anatomy; any provider trained on POCUS knows that this is a difficult and relatively non-intuitive task. 

One example of intelligent algorithms expediting this cumbersome POCUS learning process is GE’s Caption Guidance™ system. One study from JAMA Cardiology found that, “98.3% of ultrasound examinations performed by trained healthcare professionals with AI guidance were of sufficient quality to meet diagnostic standards and were not statistically different compared with images acquired by expert sonographers without AI guidance” [1].

Another study by Speranza et al. explored the ThinkSono Guidance System and discovered that it was able to guide non-expert operators through a two-point compression ultrasound technique to acquire images adequate to diagnose deep vein thrombosis often in under 10 minutes [2].

ML/DL algorithms have further expedited novice sonographer training by automating various time-consuming clinical calculations which normally require some level of expertise. Varudo et al. published a study in Critical Care assessing the clinical utility of a ML-based algorithm created for calculating real-time left ventricular ejection fraction and discovered that the “ML-enabled real-time measurements of LVEF were strongly correlated with manual measurements obtained by experts, that accuracy was excellent, precision was fair, and that reproducibility of LVEF measurements was better” [3]. Similar results have been reported for ML/DL-automating efforts for clinical measurements including IVC collapsibility [4], B-Lines [5], and a number of others.

Changes in Accessibility

As POCUS becomes more intelligent, it will also likely become increasingly more accessible. Automation of key POCUS components - including image acquisition and interpretation as discussed above - may expedite the training process of novice sonographers, a skillset in high clinical demand across specialties. This will be of particular importance in low-resource settings where ultrasound experts are scarce and training opportunities are limited. A 2024 scoping review by Kim et al. explored how ML/DL algorithms are being applied to POCUS in low-resource settings and found that while artificially intelligent POCUS can improve access and consistency where expert sonographers are scarce, barriers still remain. Notable limitations include but are not limited to low generalizability of data due to regional disparities in research, ethical challenges in remote settings, and a lack of standardized POCUS protocols, algorithms, and devices” [6].

Take-aways

ML/DL-powered POCUS can:

  • Assist novice users with image acquisition and interpretation

  • Automate clinical calculations to increase diagnostic speed and accuracy

  • Democratize POCUS to low-resource settings 

But:

  • Challenges remain and concerted effort is required to generalize algorithms and standardize protocols to realize the full potential of this technology


AUTHOR: Brady McCallister is a third-year medical student at The Warren Alpert Medical School of Brown Emergency.

FACULTY REVIEWER: Kristin Dwyer is an attending physician and clinician educator at Brown Emergency Medicine.


references

  1. Baloescu C, Bailitz J, Cheema B, et al. Artificial Intelligence–Guided Lung Ultrasound by Nonexperts. JAMA Cardiol. 2025;10(3):245–253. doi:10.1001/jamacardio.2024.4991

  2. Speranza, G., Mischkewitz, S., Al-Noor, F. et al. Value of clinical review for AI-guided deep vein thrombosis diagnosis with ultrasound imaging by non-expert operators. npj Digit. Med. 8, 135 (2025). https://doi.org/10.1038/s41746-025-01518-0

  3. Varudo R, Gonzalez FA, Leote J, Martins C, Bacariza J, Fernandes A, Michard F. Machine learning for the real-time assessment of left ventricular ejection fraction in critically ill patients: a bedside evaluation by novices and experts in echocardiography. Crit Care. 2022 Dec 14;26(1):386. doi: 10.1186/s13054-022-04269-6. PMID: 36517906; PMCID: PMC9749290.

  4. Blaivas M, Adhikari S, Savitsky EA, Blaivas LN, Liu YT. Artificial intelligence versus expert: a comparison of rapid visual inferior vena cava collapsibility assessment between POCUS experts and a deep learning algorithm. J Am Coll Emerg Physicians Open. 2020 Jul 31;1(5):857-864. doi: 10.1002/emp2.12206. PMID: 33145532; PMCID: PMC7593461.

  5. Short J, Acebes C, Rodriguez-de-Lema G, La Paglia GMC, Pavón M, Sánchez-Pernaute O, Vazquez JC, Romero-Bueno F, Garrido J, Naredo E. Visual versus automatic ultrasound scoring of lung B-lines: reliability and consistency between systems. Med Ultrason. 2019 Feb 17;21(1):45-49. doi: 10.11152/mu-1885. PMID: 30779830.

  6. Kim S, Fischetti C, Guy M, Hsu E, Fox J, Young SD. Artificial Intelligence (AI) Applications for Point of Care Ultrasound (POCUS) in Low-Resource Settings: A Scoping Review. Diagnostics (Basel). 2024 Aug 1;14(15):1669. doi: 10.3390/diagnostics14151669. PMID: 39125545; PMCID: PMC11312308.

acknowledgment:

Portions of this post were drafted and refined using ChatGPT, a language model developed by OpenAI, to support clarity and structure. All sources cited for factual content are independently verified.