Artificial Intelligence-Assisted POCUS: Promises, Pitfalls, and Path Forward

Sumit Kapoor, MD MS FCCP1*; Paul M. Wallach, MD2; Annie Whitman, RDMS3; Emily Lovallo, MD4

(1) Associate Professor, Department of Critical Care Medicine, University of Pittsburgh, Pittsburgh, PA, USA

(2) Professor of Medicine, Executive Vice Dean for Academic Affairs, University of Pittsburgh, Pittsburgh, PA, USA

(3) Sonographer Educator, Center for POCUS Education, University of Pittsburgh, Pittsburgh, PA, USA

(4) Associate Professor, Department of Emergency Medicine, University of Pittsburgh, Pittsburgh, PA, USA

*Corresponding Author: Dr. Sumit Kapoor (email: kapoors5@upmc.edu)


Download article PDF – POCUS Journal 2026;11(2):10-13

DOI: https://doi.org/10.24908/pocusj.v11i02.22263


Artificial intelligence (AI) is a transformative technology revolutionizing medical practice across specialties, and medical imaging is no exception [1]. Medical imaging has long been fertile ground for AI. At its core, imaging is a pattern-recognition problem, and machines can detect visual patterns that the human eye cannot reliably perceive [1]. Three major developments have made AI-enabled medical imaging technically feasible and commercially attractive: greater computational power, mature deep learning algorithms, and access to large imaging datasets [2]. Because they produce highly repeatable image series, computed tomography (CT) and magnetic resonance imaging (MRI) integrate seamlessly with AI [3]. However, AI in POCUS is a relatively new medical imaging application that offers unique advantages and challenges. Most contemporary ultrasound systems, from cart-based machines to handheld devices, now include sophisticated AI-enabled algorithms. These automate measurements such as left ventricular ejection fraction (LVEF), left ventricular outflow tract velocity-time integral (LVOT VTI), inferior vena cava (IVC) diameter, B-line quantification, and bladder volume calculation. 

In this editorial, we consider three important questions. Why does AI seem a natural fit for POCUS, and what promises does it offer? What challenges and limitations must be addressed before it can be adopted responsibly and at scale? What should be our path forward?

AI in POCUS: Promises

Point of care ultrasound (POCUS) is an indispensable skill with a growing range of applications, capable of answering specific clinical questions in real time at the point of care. The workflow includes image acquisition, interpretation, and medical decision-making, all occurring together at the bedside, often by the treating clinician [4]. This makes POCUS powerful but also highly operator-dependent [4,5]. The quality of decisions after POCUS depends on the operator’s skill. This operator dependence is the central reason AI matters in POCUS. Becoming proficient in POCUS requires a deliberate, time-consuming learning path that can span months to years. Limited operator expertise and a shortage of trained POCUS faculty have long constrained POCUS adoption across specialties [5]. POCUS applications continue to expand, and demand for skilled POCUS operators continues to outpace the ability to train and supervise them. AI can bridge the POCUS expertise gap by providing novice users with real-time guidance during image acquisition and coaching on transducer position and movement to obtain optimal views [5]. AI can also improve image quality by auto-adjusting gain, depth, and time-gain compensation. For novice learners, AI can ‘upskill’ them, shorten the path to proficiency, and reduce cognitive load by shifting attention from knobology and image acquisition to answering the clinical question at hand [5].

POCUS AI can promise to provide faster, more accurate POCUS image interpretation, improving user confidence [3]. For example, novice users can quantify B-lines in pulmonary edema and screen for deep venous thrombosis [4]. POCUS AI can achieve diagnostic accuracy comparable to gold-standard sonographers and clinicians in selected applications [6–8]. Some POCUS measurements, such as LVEF and LVOT VTI, are time-consuming and error-prone, even for experienced POCUS users. AI can perform these calculations in seconds [5]. Automation produces standardized images and measurements, with better intra- and inter-observer reliability [5]. For POCUS educators, AI could transform their workflow. Automated quality assurance through image-quality scoring and competency assessment could give trainees continuous feedback and reduce the burden of manual image review [1,5]. These are among the most time-consuming tasks for faculty who run POCUS programs. Auto-labelling and auto-generating standardized reports, along with automated coding and billing support, address a persistent administrative barrier [1,5]. Many POCUS examinations are performed but are never documented or billed, so the clinical work goes unrecognized, and programs lose revenue they need to sustain themselves. When a focused, AI-supported bedside examination answers the clinical question, it can reduce the need for comprehensive ultrasound studies. It can also ease the workload of busy echocardiography and radiology departments and shorten time to diagnosis [5].

Using POCUS AI during  could improve safety and first-pass success rates [5]. In low- and middle-income countries, POCUS AI-assisted image acquisition and interpretation could extend ultrasound to patients who would otherwise go without imaging [9]. It can ‘skill-enable’ clinicians who would not otherwise be able to use the modality. Looking further ahead, AI-guided patient self-scanning may support home-based monitoring [1]. One could imagine serial self- lung or IVC assessments by patients with heart failure or kidney disease.

The role of AI in POCUS is therefore evolving in two directions simultaneously. It supports novices by lowering the barrier to entry and upskilling them, and it benefits experts by automating repetitive tasks, adding consistency and reducing strain on them. Table 1 lists important POCUS AI applications.

Table 1. Artificial Intelligence (AI) applications for POCUS across five development domains.
POCUS AI Development DomainHow AI can helpWorkflowExamples
Education· Image Acquisition
· Auto gain, auto depth features
· Auto-labelling of sonographic anatomy
· Real-time feedback on transducer position and movement for image optimization
· Improves image quality and provides immediate feedback
· Help learners obtain quality cardiac views, perform deep venous thrombosis (DVT) scans, and Focused Assessment with Sonography for Trauma (FAST) exams
· Help learn sonographic anatomy
Clinical and Workflow· Image Interpretation
· Task Automation
· Procedural guidance through heads-up display and labelling of sonographic anatomy
· Detects POCUS-related abnormalities and automated calculations
· Automates measurements, auto-labelling, and auto-saving images and cine clips
· Heads-up display and labelling of sonographic anatomy
· Cardiac: left ventricular systolic function, diastolic function, valvular pathologies, pericardial effusion, pulse checks during cardiopulmonary resuscitation, right ventricular tricuspid annular plane systolic excursion (TAPSE), right ventricular fractional area change (FAC), automated measurement of left ventricular ejection fraction, velocity time integral for cardiac output calculation
· Lung: A-line and B-line detection, pneumothorax, pleural effusion, lung consolidation
· Renal/urology: automated bladder volume measurement, obstructive hydronephrosis detection
· Vascular: DVT detection, peripheral arterial disease detection, automated calculation of Inferior vena cava (IVC) collapsibility index
· Neurologic: automated optic nerve sheath Diameter measurement
· Trauma: FAST and eFAST exam
· Obstetric and Gynecologic: intrauterine and ectopic pregnancy, fetal age estimation
· Breast: suspected breast lesions
· Musculoskeletal: wrist fracture detection
· Endocrine: thyroid nodule screening
· Pediatric: hip dysplasia detection
· Identifying safe fluid pockets for drainage procedures such as thoracentesis and paracentesis, and guiding vascular access procedures, including central line, arterial line, and peripheral IV placement
Research and Administrative· Automated POCUS image classification
· Automated report generation
· Automated coding and billing
· Quality assurance (QA)
· Help with archiving images
· Help with running the POCUS program with QA, classification of POCUS images in the archived database, and creating reports and dashboards for the POCUS program
· Help with managing the POCUS database for research, including sorting images

AI in POCUS: Challenges, Limitations and Pitfalls

Although AI technology is rapidly integrating with POCUS, it remains unclear whether POCUS AI adoption guides patient management, improves patient outcomes, optimizes resource utilization, and provides overall clinical utility [10]. How it impacts workflows for POCUS learners, educators, and administrators remains unanswered [10].  Some inherent challenges of AI are also applicable to POCUS, and others are unique to POCUS. Many POCUS datasets are small, noisy, single-center, homogeneous, and derived mainly from high-quality expert-acquired images [1,3,4]. Models trained on such datasets may perform poorly in real-world patients. Bias may arise from age, sex, ethnicity, body habitus, and image quality [1,3,4]. Over time, domain shift and model drift can erode performance silently as patient populations, ultrasound devices, and practice patterns change [1,3,4]. POCUS images also pose technical interpretation challenges compared to natural images [4]. They differ fundamentally from natural images, contain artifacts, acoustic shadowing, speckle, indistinct tissue boundaries, and change with every adjustment of probe angle, pressure, and machine settings. This makes training and labelling on POCUS datasets challenging. The success of any AI tool depends heavily on the quality of input data; ‘garbage in, garbage out’ could be a problem with poor-quality POCUS datasets.

AI interpretation could become a problem especially in critically ill patients who are difficult to position, mechanically ventilated, edematous, or obese, and in whom diagnostic-quality windows are hard to obtain [3]. Most AI outputs serve as ‘black boxes’ without clear explanation of how they were generated [1,3,4]. This lack of transparency undermines clinician trust. The problem is compounded by the proliferation of proprietary, vendor-specific algorithms. There is a risk of automation bias: uncritical acceptance of an algorithm’s output [1,4]. For novice users, overreliance and blind trust may lead to de-skilling, mis-skilling, or never-skilling [1,4]. Many AI tools are poorly integrated into clinical workflows and electronic health records, limiting their practical use [10]. Regulatory pathways for AI algorithms are still maturing [3,10]. Protecting patient privacy is essential when models are developed using images. When an AI-guided decision contributes to an undesired outcome, accountability is unclear: the clinician, the institution, or the AI developer [10]. Figure 1 lists major promises and pitfalls of POCUS AI.

Figure 1. Major promises and pitfalls of Artificial Intelligence (AI) in POCUS.

Ferre and colleagues identified major barriers to adopting POCUS AI [10]. Most AI tools are technology-driven rather than outcome-driven solutions that focus on solving real-world problems. Data limitations, poor integration into existing clinical workflows, and unresolved regulatory and ethical concerns were additional barriers.

In 2025, an international web-based cross-sectional survey of 1,154 healthcare professionals assessed familiarity with, perceptions of, and adoption of AI in POCUS [2]. Most respondents were experienced emergency medicine and critical care clinicians who use POCUS in their daily work routines. Yet 77% did not use any AI- or machine learning–assisted tool in patient care. Most believed AI could expedite diagnosis, improve accuracy and confidence, and reduce inter-operator variability. However, they cited concerns about inadequate training resources, lack of credentialing pathways, doubts about workflow integration, and limited trust. Many said they would need to verify AI findings themselves before acting on them. The lack of guidelines or endorsement from professional societies, along with cultural resistance to new technology, also slowed adoption. Training and education, followed by clinical validation and evidence, emerged as the leading barriers. The ‘black box’ nature of AI was a key driver of mistrust, and regulatory approval and clinical validation were seen as solutions to build trust. Notably, enthusiasm for AI in POCUS remained high.

The Path Forward

AI in POCUS is a promising yet still-maturing field. Whether its adoption improves patient outcomes, optimizes resource use, or meaningfully guides management remains to be seen. Realizing its potential will require deliberate effort from all stakeholders. Ferre et al. proposed a framework to identify opportunities for POCUS AI development across five domains: education, clinical, workflow, research, and administration [10]. They also developed an 8-step process to improve POCUS AI adoption across these five domains.

Industry should build POCUS AI tools around end users and real clinical problems [4,10]. Developers and researchers need access to large, diverse, multicenter datasets that include difficult and poor-quality images [4,10]. AI models should be transparent and explainable, validated across vendors, and monitored after deployment for drift [4,10]. Educators and professional societies also play an important role. Curricula should teach clinicians how AI tools work, their limitations, when they can fail, and how to verify their outputs [10]. Professional medical societies should develop training and credentialing standards for POCUS AI, and government agencies should issue regulatory approvals [10]. Trainees should develop independent hands-on acquisition and interpretation skills so that AI augments competence and clinical decision-making rather than replacing them.

AI is a powerful tool that can increase POCUS adoption and empower clinicians. It is a path toward precision medicine and, eventually, agentic or autonomous systems that integrate multi-organ findings. No matter how sophisticated AI algorithms become, human oversight, critical thinking, sound education, and, above all, clinical context remain indispensable. AI may change how we hold the probe, but it should not change who makes clinical decisions and who is responsible for the patient.

Disclosure Statement

The authors have no conflicts of interest to disclose.

References

1. Kayarian F, Patel D, O’Brien JR, Schraft EK, Gottlieb M. Artificial intelligence and point-of-care ultrasound: Benefits, limitations, and implications for the future. Am J Emerg Med. 2024;80:119-122. doi:10.1016/j.ajem.2024.03.023

2. Wong A, Roslan NL, McDonald R, Noor J, Hutchings S, D’Costa P, Via G, Corradi F. Clinical obstacles to machine-learning POCUS adoption and system-wide AI implementation (The COMPASS-AI survey). Ultrasound J. 2025;17(1):32. doi: 10.1186/s13089-025-00436-2

3. Kim J, Maranna S, Watson C, Parange N. A scoping review on the integration of artificial intelligence in point-of-care ultrasound: Current clinical applications. Am J Emerg Med. 2025;92:172-181. doi:10.1016/j.ajem.2025.03.029

4. Vega R, Dehghan M, Nagdev A, Buchanan B, Kapur J, Jaremko JL, Zonoobi D. Overcoming barriers in the use of artificial intelligence in point of care ultrasound. NPJ Digit Med. 2025;8(1):213. doi: 10.1038/s41746-025-01633-y

5. Wu D, Arntfield R, Millington SJ. Artificial Intelligence in Point-of-Care Ultrasound. J Intensive Care Med. 2026:8850666261427329. doi: 10.1177/08850666261427329

6. Stringer JSA, Pokaprakarn T, Prieto JC, Vwalika B, Chari SV, Sindano N, Freeman BL, Sikapande B, Davis NM, Sebastião YV, Mandona NM, Stringer EM, Benabdelkader C, Mungole M, Kapilya FM, Almnini N, Diaz AN, Fecteau BA, Kosorok MR, Cole SR, Kasaro MP. Diagnostic Accuracy of an Integrated AI Tool to Estimate Gestational Age From Blind Ultrasound Sweeps. JAMA. 2024;332(8):649-657. doi: 10.1001/jama.2024.10770

7. Asch FM, Mor-Avi V, Rubenson D, Goldstein S, Saric M, Mikati I, Surette S, Chaudhry A, Poilvert N, Hong H, Horowitz R, Park D, Diaz-Gomez JL, Boesch B, Nikravan S, Liu RB, Philips C, Thomas JD, Martin RP, Lang RM. Deep Learning-Based Automated Echocardiographic Quantification of Left Ventricular Ejection Fraction: A Point-of-Care Solution. Circ Cardiovasc Imaging. 2021;14(6):e012293. doi: 10.1161/CIRCIMAGING.120.012293

8. Libon J, Ng C, Bailey A, Hareendranathan A, Joseph R, Dulai S. Remote diagnostic imaging using artificial intelligence for diagnosing hip dysplasia in infants: Results from a mixed-methods feasibility pilot study. Paediatr Child Health. 2023;28(5):285-290. doi:10.1093/pch/pxad013

9. 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;14(15):1669. doi:10.3390/diagnostics14151669

10. Ferre RM, Liu RB, Lam HS, Blaivas M, Oh L, Sangal RB, Baloescu C, Adhikari S. Artificial Intelligence in Point-of-Care Ultrasound: Domains, Barriers and a Framework for Future Development. J Am Coll Emerg Physicians Open. 2026;7(5):100491. doi: 10.1016/j.acepjo.2026.100491

Posted in .

Leave a Reply

Your email address will not be published. Required fields are marked *