A prospective study has demonstrated the potential feasibility, applicability and accuracy held by artificial intelligence (AI) in the detection of carotid artery disease on greyscale static duplex ultrasound imaging. This is the conclusion reached by Ali Kordzadeh (Faculty of Science and Engineering, Anglia Ruskin University, Chelmsford, UK) and colleagues in the journal Vascular.
“This network has the potential to be used as a standalone software or to be embedded in any DUS [duplex ultrasound] machine,” they add. “This can enhance carotid artery disease recognition with limited or no vascular experience, or serve as a stratification tool for tertiary referral, further imaging and overall management.”
To evaluate the feasibility, applicability and accuracy of AI in the detection of normal versus carotid artery disease through greyscale static duplex ultrasound images, Kordzadeh et al conducted a prospective image acquisition of individuals undergoing duplex sonography for suspected carotid artery disease at a single hospital (Broomfield Hospital, Mid and South Essex NHS Foundation Trust, Chelmsford, UK).
A total of 156 images of normal and stenotic carotid arteries—based on North American Symptomatic Carotid Endarterectomy Trial (NASCET) criteria—were evaluated using geometry group network based on convolutional neural network (CNN) architecture. Outcomes were reported based on sensitivity, specificity and accuracy of the AI network for detecting normal versus stenotic carotid arteries, as well as various categories of carotid artery stenosis.
Detailing their results, Kordzadeh et al state that the overall sensitivity, specificity and accuracy of AI was 91%, 86% and 92%, respectively, in the detection of normal carotid artery stenosis, and 87%, 82% and 90%, respectively, for any carotid artery stenosis.
In addition, subgroup analyses demonstrated that the AI network in question has the ability to detect stenotic carotid artery images (<50%) versus normal carotid artery images with a sensitivity of 92%, a specificity of 87% and an accuracy of 94%. The authors further note that these values (sensitivity, specificity and accuracy) for a subgroup of 50–75% stenosis versus normal were 84%, 80% and 88%, respectively, and—for another subgroup with carotid artery disease with >75% stenosis—were 90%, 83% and 92%, respectively.
“The biggest advantage of AI is embedded in moving away from qualitative analysis (operator dependent) to a quantitative, data-driven approach in detecting carotid artery stenosis,” said Kordzadeh, speaking to NeuroNews following the publication of these findings. “AI has the potential to reduce human error by a significant margin with [an] ongoing data feed that continues to evolve for better accuracy, even in detection of stable or unstable plaque in addition to carotid artery stenosis. This enhances carotid artery disease recognition in a timely manner, resulting in prompt management, and reductions in mortality and morbidity associated with transient ischaemic attacks or stroke in the carotid artery territory.”