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).
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).
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).
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).
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).
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).
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).
Power, M., Fell, G. & Wright, M. Principles for high-quality, high-value testing. BMJ Evid.-Based Med. 18, 5–10 (2013).
Wong, H. J. & Morra, D. Excellent hospital care for all: open and operating 24/7. J. Gen. Intern. Med. 26, 1050–1052 (2011).
Stein, P. D. et al. Venous thromboembolism according to age: the impact of an aging population. Arch. Intern. Med. 164, 2260–2265 (2004).
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).
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).
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).
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).
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
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
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/.
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
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41746-025-01613-2
