In a prospective, controlled clinical study with deep vein thrombosis (DVT) triage with Artificial Intelligence (AI)-guided software for simulating compression ultrasound, lead investigator Efthymios Avgerinos (University of Athens, Athens, Greece) and his colleagues demonstrated high sensitivity and specificity in DVT diagnosis. The researchers concluded that the machine learning software was able to help non-experts obtain valid ultrasound images of venous compressions and enabled safe and efficient remote DVT triage. Avgerinos recently presented these results at the 22nd Annual Meeting of the European Vein Forum (EVF). (June 30–July 2, 2022, Venice, Italy).
Diagnostic algorithms for DVT typically require a clinical probability test assessment, standard lower leg duplex or at least two-point compression ultrasound, and a D-dimer test. These can be time-consuming for both patients and doctors in an otherwise busy emergency room or vascular ultrasound lab, Avgerinos said Venous Messages.
AutoDVT (ThinkSono), a machine learning software, is an AI tool that guides laypersons in acquiring appropriate two-point compression sequences (common femoral and popliteal veins). Speaking at EVF, Avgerinos explained that a smartphone or tablet and a wireless ultrasound probe are the only devices needed to use this technology, and that users upload captured images to the cloud, which are then reviewed remotely by an expert.
In Avgerinos et al‘s study, researchers recruited patients with suspected DVT at two tertiary centers: Magdeburg, Germany and Athens, Greece. Enrolled patients underwent an AutoDVT scan by a non-sonographer prior to standard duplex scanning. Two to four external qualified doctors blindly reviewed images collected by the software and uploaded to a cloud-based platform. Based solely on these images, all reviewers rated all sequences on the American College of Emergency Physicians (ACEP) image quality scale (score 1-5, where a rating of three or higher is defined as adequate diagnostic quality) and made a triage decision: low or high risk of DVT. The categorization was compared to the DVT diagnosis by the standard duplex scan.
Avgerinos addressed the EVF audience and reported that three nurses treated 37 patients (age 63.7 ± 17.01, body mass index [BMI] 28.62 ± 5.9, 32% females), resulting in 34 (97%) scans judged to be of diagnostic quality. The mean ACEP scores were 3.88 ± 0.43, and there was no significant difference in the ratio of diagnostic quality of the scans between examiners or scanners, the moderator shared.
In addition, Avgerinos reported that AutoDVT triaged 23 (62%) scans as negative, all of which the standard duplex scan confirmed as negative. Of the 14 patients classified as high-risk, seven were positive for DVT. These results explained a sensitivity of 100% and a specificity of 77% for DVT diagnosis.
Speak with Venous Messages After the presentation, Avgerinos summarized the key findings of the study: “The machine learning software was able to help non-experts obtain valid ultrasound images of venous compressions and enabled safe and efficient remote triage. Given that the vast majority of requested CBCT scans are negative, such a triaging strategy allows for faster diagnosis and treatment of high-risk patients and can save the need and cost of multiple unnecessary duplex scans. Patient waiting times can be reduced and radiologist and sonographer resources reallocated.”
“ThinkSono investigators look forward to evaluating these results on a larger scale,” the moderator added, noting that several European sites are integrating this new technology and commercial pilots are underway.