Artificial intelligence guided imaging as a tool to fill gaps in health care delivery

Deep vein thrombosis (DVT) causes significant morbidity/mortality and timely diagnosis often via ultrasound is critical. However, the shortage of trained ultrasound providers has been an

ongoing challenge. Recently, Speranza and colleagues (2025) demonstrated that an artificial intelligence (AI) guided ultrasound system used by non-ultrasound-trained nurses with remote

clinician review can achieve sensitivities of 90–98% and specificities of 74–100% for diagnosing DVT. This study highlights the potential for AI guided imaging to address important gaps in

health care delivery. INTRODUCTION Deep vein thrombosis (DVT) involves blood clot formation primarily within major leg veins, which can travel to the lungs, leading to life-threatening

pulmonary embolism1. Therefore, prompt diagnosis and management of DVT is critical1. Although invasive venography is the gold standard, the most common way to diagnose DVT is through

ultrasound, whereby a technician uses an ultrasound probe to compress the vein(s) with suspected DVT2. A non-compressible vein is diagnostic for DVT, and treatment is generally started with

blood thinners2. Most clinicians are not trained to perform DVT ultrasounds, and therefore, ultrasound technicians are necessary3; however, they are not always available4. A survey of 79 US

teaching hospitals showed that only 24% had ultrasound technicians on the premises after hours between 6pm and 8am, when patients can present acutely with DVT’s5. Although other imaging

modalities can be used to diagnose DVT, they are generally more expensive and can be less accurate than ultrasound6. Empiric treatment of suspected DVT’s with blood thinners prior to imaging

confirmation carries considerable risks, including life-threatening bleeding7. Therefore, there is an important need to address this gap in diagnostic care. Artificial intelligence (AI) has

the potential to guide non-ultrasound-trained providers in performing DVT ultrasounds with sufficient diagnostic quality8. Recently, Speranza and colleagues (2025) assessed the use of

AI-guided ultrasound by non-ultrasound-trained nurses to diagnose DVT9. In this article, we highlight key findings from Speranza et al.’s study and discuss the potential for AI-guided

imaging to fill important gaps in healthcare delivery. AN AI-GUIDED ULTRASOUND SYSTEM Speranza and colleagues evaluated the _ThinkSono Guidance System_ (ThinkSono GmbH;

https://thinksono.com/), an AI system that guides non-ultrasound-trained providers in performing DVT ultrasounds9. The system consists of a smartphone application and a portable ultrasound

device9. The software guides the operator in using the ultrasound probe to compress the common femoral and popliteal veins (the highest probability sites for identifying DVT’s) several

times1,9. Each time, the operator receives feedback on the compression location, timing, and positioning9. The entire process generally takes less than 10 minutes9. Once the software

confirms the acquisition of appropriate images, they are uploaded to a secure cloud dashboard and reviewed by a remote medical imaging specialist to diagnose DVT9. KEY FINDINGS BY SPERANZA

AND COLLEAGUES Speranza and colleagues conducted a retrospective analysis of data from a multicenter, prospective, double-blinded pilot study designed to evaluate the accuracy of AI-guided

ultrasound imaging for DVT diagnosis9. The study was conducted across 11 UK hospitals and included adults with symptoms suggestive of DVT who required a diagnostic ultrasound scan9. Overall,

381 patients between 2021–2023 were included, all of whom underwent both an AI-guided ultrasound scan by a non-ultrasound-trained nurse and an expert sonographer-performed ultrasound scan9.

Each AI-guided scan was reviewed remotely by a blinded radiologist or point-of-care ultrasound (POCUS) certified emergency physician9. The DVT diagnosis made by the remote clinician using

images from the AI-guided ultrasound was compared to the ground truth, defined as the DVT diagnosis obtained from the local imaging specialist report based on the expert sonographer scan,

which is the standard of care9. In the study, 80% of AI-guided scans received an American College of Emergency Physicians (ACEP) image quality score ≥ 3, which constitutes adequate image

quality to diagnose DVT9. The AI-guided ultrasound system achieved sensitivities of 90-98% and specificities of 74–100% for diagnosing DVT9, which can be considered a good to excellent

diagnostic test as defined by Power and colleagues (2013)10. IMPLICATIONS OF AN AI-GUIDED ULTRASOUND SYSTEM TO DIAGNOSE DVT Speranza and colleagues demonstrated that an AI-guided ultrasound

system used by non-ultrasound-trained nurses with remote clinician review can achieve relatively high sensitivity and specificity for diagnosing DVT9. Since nurses are generally available

24/7 in hospital settings, this technology has the potential for important clinical impact11. Given our aging and increasingly sedentary society with a rising incidence of DVT’s in

combination with constrained health care budgets and worker shortages, the lack of available ultrasound technicians to perform DVT scans will be an escalating challenge12,13. This technology

could temporarily address the shortage of trained ultrasound providers while long-term solutions are implemented, including increasing health care funding, worker capacity, and

infrastructure14. The other potential application of AI-guided ultrasound systems is the ability to reduce operator variability, even amongst expert sonographers15. Ultrasound scanning can

vary significantly between providers based on the specific location scanned, probe angle, and amount of pressure applied, among other factors16. This variability can lead to differences in

image quality and inaccurate diagnoses16. AI-guided systems can make the scanning process more systematic through real-time instructions and feedback, thereby reducing variability amongst

operators15. LIMITATIONS Although the performance metrics of the AI-guided ultrasound system for diagnosing DVT are satisfactory, there is room for improvement9. Based on the study findings,

20% of AI-guided scans would be of insufficient quality, and a non-negligible proportion of DVT diagnoses would be incorrect, leading to inappropriate management9. Therefore, additional

refinement of the technology to improve image quality and diagnostic accuracy may further increase its clinical utility9. Additionally, this was a retrospective analysis of data from a pilot

study9. Prospective validation of this technology on larger cohorts is needed to support clinical implementation. CONCLUSIONS Through a retrospective analysis of a multicenter, prospective,

double-blinded pilot study, Speranza and colleagues (2025) demonstrated that an AI-guided ultrasound system used by non-ultrasound-trained nurses with remote clinician review can achieve

relatively high sensitivity and specificity for diagnosing DVT9. This study highlights the potential for AI tools to fill important gaps in healthcare delivery by supporting timely DVT

diagnosis when ultrasound technicians are not available9. Importantly, AI-guided imaging can be applied to many other areas of health care to standardize imaging protocols, reduce operator

variability, and potentially improve diagnostic accuracy, ultimately improving patient care. DATA AVAILABILITY No datasets were generated or analyzed during the current study. REFERENCES *

Waheed, S. M., Kudaravalli, P. & Hotwagner, D. T. Deep Vein Thrombosis. in _StatPearls_ (StatPearls Publishing, Treasure Island (FL), 2025). * Baker, M., Anjum, F. & dela Cruz, J.

Deep Venous Thrombosis Ultrasound Evaluation. in _StatPearls_ (StatPearls Publishing, Treasure Island (FL), 2025). * Russell, F. M. et al. The state of point-of-care ultrasound training in

undergraduate medical education: findings from a national survey. _Acad. Med. J. Assoc. Am. Med. Coll._ 97, 723–727 (2022). Article  Google Scholar  * Tay, E. T., Stone, M. B. & Tsung,

J. W. Emergency ultrasound diagnosis of deep venous thrombosis in the pediatric emergency department: a case series. _Pediatr. Emerg. Care_ 28, 90–95 (2012). Article  PubMed  Google Scholar

  * Desser, T. S., Rubin, D. L. & Schraedley-Desmond, P. Coverage of emergency after-hours ultrasound cases: survey of practices at U.S. Teaching hospitals. _Acad. Radiol._ 13, 249–253

(2006). Article  PubMed  Google Scholar  * Garcia-Bolado, A. & Del Cura, J. L. CT venography vs ultrasound in the diagnosis of thromboembolic disease in patients with clinical suspicion

of pulmonary embolism. _Emerg. Radiol._ 14, 403–409 (2007). Article  PubMed  Google Scholar  * Eisenson, D. L. et al. Prevalence and consequences of empiric anticoagulation for venous

thromboembolism in patients hospitalized for COVID-19: a cautionary tale. _J. Thromb. Thrombolysis_ 52, 1056–1060 (2021). Article  CAS  PubMed  PubMed Central  Google Scholar  * Nothnagel,

K. & Aslam, M. F. Evaluating the benefits of machine learning for diagnosing deep vein thrombosis compared with gold standard ultrasound: a feasibility study. _BJGP Open_ 8,

BJGPO.2024.0057 (2025). Article  PubMed  Google Scholar  * Speranza, G., Mischkewitz, S., Al-Noor, F. & Kainz, B. Value of clinical review for AI-guided deep vein thrombosis diagnosis

with ultrasound imaging by non-expert operators. _Npj Digit. Med._ 8, 1–7 (2025). Article  Google Scholar  * Power, M., Fell, G. & Wright, M. Principles for high-quality, high-value

testing. _BMJ Evid.-Based Med._ 18, 5–10 (2013). Article  Google Scholar  * Wong, H. J. & Morra, D. Excellent hospital care for all: open and operating 24/7. _J. Gen. Intern. Med._ 26,

1050–1052 (2011). Article  PubMed  PubMed Central  Google Scholar  * Stein, P. D. et al. Venous thromboembolism according to age: the impact of an aging population. _Arch. Intern. Med._ 164,

2260–2265 (2004). Article  PubMed  Google Scholar  * Won, D. et al. Sound the alarm: The Sonographer shortage is echoing across healthcare. _J. Ultrasound Med. Off. J. Am. Inst. Ultrasound

Med._ 43, 1289–1301 (2024). Google Scholar  * Agyeman-Manu, K. et al. Prioritising the health and care workforce shortage: protect, invest, together. _Lancet Glob. Health._ 11, e1162–e1164,

https://doi.org/10.1016/S2214-109X(23)00224-3 (2023). Article  CAS  PubMed  Google Scholar  * Edwards, C., Chamunyonga, C., Searle, B. & Reddan, T. The application of artificial

intelligence in the sonography profession: Professional and educational considerations. _Ultrasound J. Br. Med. Ultrasound Soc._ 30, 273–282 (2022). Google Scholar  * Lee, H. J. et al.

Intraobserver and interobserver variability in ultrasound measurements of thyroid nodules. _J. Ultrasound Med. Off. J. Am. Inst. Ultrasound Med._ 37, 173–178 (2018). Google Scholar  Download

references ACKNOWLEDGEMENTS This editorial did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. AUTHOR INFORMATION AUTHORS AND

AFFILIATIONS * Division of Vascular Surgery, University of Toronto, Toronto, ON, Canada Ben Li * Temerty Centre for Artificial Intelligence Research and Education in Medicine, University of

Toronto, Toronto, ON, Canada Ben Li * Harvard Medical School, Boston, MA, USA Elizabeth J. Enichen, Kimia Heydari & Joseph C. Kvedar Authors * Ben Li View author publications You can

also search for this author inPubMed Google Scholar * Elizabeth J. Enichen View author publications You can also search for this author inPubMed Google Scholar * Kimia Heydari View author

publications You can also search for this author inPubMed Google Scholar * Joseph C. Kvedar View author publications You can also search for this author inPubMed Google Scholar CONTRIBUTIONS

B.L. wrote the first draft of the manuscript. E.J.E., K.H., and J.C.K. provided critical revisions. All authors have read and approved of the final manuscript. CORRESPONDING AUTHOR

Correspondence to Ben Li. ETHICS DECLARATIONS COMPETING INTERESTS J.C.K. is the editor-in-chief of _npj Digital Medicine_. All other authors declare no competing interests. RIGHTS AND

PERMISSIONS OPEN ACCESS This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any

medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The

images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not

included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly

from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. Reprints and permissions ABOUT THIS ARTICLE CITE THIS ARTICLE Li, B., Enichen,

E.J., Heydari, K. _et al._ Artificial intelligence guided imaging as a tool to fill gaps in health care delivery. _npj Digit. Med._ 8, 248 (2025). https://doi.org/10.1038/s41746-025-01613-2

Download citation * Received: 04 March 2025 * Accepted: 07 April 2025 * Published: 05 May 2025 * DOI: https://doi.org/10.1038/s41746-025-01613-2 SHARE THIS ARTICLE Anyone you share the

following link with will be able to read this content: Get shareable link Sorry, a shareable link is not currently available for this article. Copy to clipboard Provided by the Springer

Nature SharedIt content-sharing initiative