“Unveiling the invisible: Harnessing AI for carotid web diagnosis—A futuristic approach to enhance vascular healthcare”

We read with great interest “Multimodality imaging of carotid web: A case report and literature review” by Zhang et al. ¹ The authors presented a unique case study of a 38-year-old man with a carotid web unidentified in the initial report, retrospectively confirmed using digital subtraction angiography and histopathology. Through a literature review of 80 reports, including 681 patients, the authors recommended multimodality imaging for carotid webs (CWs).

We agree with the authors’ conclusion that CWs are a rare cause of ischemic stroke, and they are easily underdiagnosed or misdiagnosed. We propose using artificial intelligence (AI) to reduce diagnostic challenges. Artificial intelligence is becoming ubiquitous within medicine, especially in diagnostics. Deep convolutional neural networks (DCNNs) have been demonstrated to be monumental in electrocardiogram interpretation. They allow a “human-like” interpretation of ECGs, closely identifying cardiac diseases, such as left ventricular (LV) systolic dysfunction, hypertrophic cardiomyopathy (HCM), and silent atrial fibrillation (AF). ² Interestingly, 20% of patients with silent atrial fibrillation (AF) are asymptomatic. Another one-third of patients have atypical symptoms. Less than 50% of patients have typical symptoms. ³

To identify silent AF, a group from the Mayo Clinic developed a DCNN for electrocardiogram. Using a dataset of ECGs from 126,526 patients, they built the DCNN to identify silent AF. ² The program had a sensitivity of 79.0%, a specificity of 79.5%, and an accuracy of 79.4% in detecting patients with documentation of AF using only the sinus-rhythm electrocardiogram. ²

As CW diagnosis is elusive, akin to silent AFs, we propose a similar development of a DCNN to address diagnostic challenges. Mellander et al. (2023) evaluated the ability of AI software to detect large vessel occlusions in the brain using computed tomography angiography. ⁴ They reported a sensitivity of 84% for T-type internal carotid occlusions. ⁴ We believe a similar program can be used as an “AI diagnostic check” integrated into the patient’s EHR. Algorithms can be fed demographics, past medical history, medications, and initial raw diagnostic imaging to run DCNNs. The testing would be primarily for all elusive diagnostic radiology entities, such as a CW. The findings would be summarized and presented to the patient’s physician with numerical probabilities, aiding the diagnosis of CWs. Asymptomatic CWs may be monitored, while symptomatic CWs may be swiftly addressed with appropriate treatment. The DCNN may also be programmed to intake retrospective scans, enhancing asymptomatic monitoring abilities and computing probabilities of symptomatic onset timing.

Developing and using such novel AI technology in vascular practices will decrease the likelihood of missing a CW diagnosis, bolstering positive patient outcomes. Its mere presence in the “AI diagnostic check” complimentary EHR program may also educate physicians who have never learned about CWs to become aware of its etiologic potential in cryptogenic stroke.