Keywords:
Artificial Intelligence, eHealth, Vascular, Ultrasound, Diagnostic procedure, Embolism / Thrombosis
Authors:
J. Oppenheimer, R. Mandegaran, B. Kainz, M. P. Heinrich, F. Noor, S. Mischkewitz, A. Ruttloff, P. Klein-Weigel
DOI:
10.26044/ecr2022/C-10357
Methods and materials
Patients with a suspected DVT were recruited at a German tertiary care clinic via the emergency department or primary care referral. Each patient underwent two DVT examinations. The first was performed by one of two medical technicians without formal DVT-ultrasound experience or training. Subsequently each patient underwent a full leg ultrasound examination by a highly-trained angiology consultant with 26 years of experience with DVT ultrasound diagnosis.
During the diagnostic scan, the leg was scanned with multiple compressions along the veins from the inguinal crease to the popliteal fossa and continuing on to the calf, with at least one compression on each pair of calf veins. The specialist was able to use any ultrasound device provided by their institution and only the final diagnosis for proximal DVT (up to and including the popliteal fossa) was recorded. The technician used a handheld, wireless linear ultrasound device (Clarius L7 HD, Clarius Mobile Health Corp., Vancouver, Canada) in conjunction with a proprietary app-based image-acquisition software, AutoDVT. AutoDVT instructs users to find appropriate vessel landmarks on the legs and perform compression sequences at these points with the help of machine-learning-based algorithms (figures 1-3). The software instructs users through a three-point proximal DVT detection method as follows:
- Two compressions are performed in the groin area: (i) of the veins at the saphenofemoral junction, and (ii) of the common femoral vein.
- One compression is performed of the superficial femoral vein at the level of the thigh.
- Two compressions are performed in the popliteal fossa of the popliteal vein.
Scans with at least one complete compression sequence of each point were marked as complete 3-point-compression scans, additionally scans with at least one complete sequence in the groin and knee area were noted as complete 2-point-compression scans, reflecting variable clinical protocols in use [3]. Complete scans collected by the software were then uploaded to a cloud-based platform (figures 4-6) for review by an external board-certified radiologist with nine years of experience and a final-year medical student with four years of training in ultrasound DVT-diagnosis. Both reviewers made a diagnosis based on these images only, and rated each compression sequence as well as the full scan on the 5-point American College of Emergency Physicians (ACEP) Emergency Ultrasound Reporting Guidelines image quality scale (see figure 7 - table 1) [4]. Full scans with an ACEP score below 3 were defined as non-diagnostic. Sensitivity and specificity were defined in comparison to the results of the compression ultrasound performed by the specialist on site. For analysis of image quality of 2-point-compression only, an additional subgroup analysis of ACEP scores was performed on a per compression basis for only groin and knee compressions and the medians (rounded to the nearest full number, except in calculation of means) of these scores was defined as 2-point-compression ACEP scores. Results are displayed as means with 95% confidence intervals (±) to better convey relevant differences in scoring.