Artificial Intelligence, Telemedicine, Information Quality, Global Vascular Surgery, and Future Directions

72.1Artificial intelligence readiness and reporting

Mature vascular artificial intelligence applications assist with image classification, segmentation, structured measurement, case identification, triage, or risk stratification, leaving the final clinical action to a clinician Li 2022 · Stonko 2023. Autonomous operative capabilities and unsupervised bedside decision-making remain experimental Liyanage 2026 · Abara 2025 · Kiang 2024. Vascular artificial intelligence tools function as decision support to narrow attention or standardise measurement, and they do not remove clinical accountability.

Model deployment requires three distinct layers of validation:

• Technical validity: the model performs accurately on data resembling the intended local patients, scanners, and clinical records.
• Clinical validity: the output aligns with a genuine decision point, such as urgent assessment, repeat imaging, or multidisciplinary discussion.
• Workflow validity: the result arrives at the correct time, to the correct person, in an actionable form without causing silent delay, duplicated work, or alarm fatigue.

For EU medical-device AI, Article 6(1) high-risk classification requires both a relevant Annex I product or safety-component role and a requirement for third-party conformity assessment under the applicable product legislation. Not every clinical AI application is automatically high-risk AIB 2025 · Regulation 2024.

As at 20 September 2026, Regulation (EU) 2026/1744 has amended the timetable: Chapter III, Sections 1–3, except Article 6(5), apply from 2 December 2027 to Article 6(2)/Annex III systems and from 2 August 2028 to Article 6(1)/Annex I systems. These dates supersede the 2 August 2027 medical-device date in the June 2025 MDCG document. Other AI Act provisions have their own application and transition rules; MDR/IVDR obligations are not suspended Regulation 2026 · AIB 2025 · Regulation 2024.

72.2Governance and lifecycle monitoring

Governance determines whether an accurate model remains safe in clinical practice. Publications from the World Health Organization (WHO) support responsible governance but are not binding regulatory requirements: the 2021 report offers ethical recommendations, while the 2023 publication is a high-level overview, not a regulatory framework or policy. The FDA's 2021 AI/ML-SaMD Action Plan is a historical programme of regulatory work, not an exhaustive statement of current duties WHO 2023 · WHO 2021 · FDA 2021.

72.3Telemedicine workflows and escalation

Telemedicine pathways are safe only when integrated with defined clinical escalation routes Yew 2023 · Montano 2022 · Lareyre 2022. Remote care is established for structured postoperative wound checks with images, selected endovascular follow-up, medication management, surveillance-result discussion, and stable symptom triage. The pathway defines which symptoms are suitable for remote assessment and which findings mandate immediate direct examination.

Absolute triggers for hands-on evaluation include:

• Enlarging wounds or new tissue loss.
• New rest pain or a cool extremity.
• Neurological symptoms following carotid intervention.
• Dialysis access dysfunction.
• Fever following graft placement.

Telemedicine programs monitor equity to ensure remote pathways do not exclude patients with limited digital literacy, poor connectivity, sensory impairment, insecure housing, or language barriers. Routine quality tracking includes missed escalations, delayed imaging, and unplanned admissions Yew 2023 · Montano 2022 · Lareyre 2022.

72.4Patient-facing information quality

Patient-facing vascular information on the internet has documented limitations in readability, usability, completeness, and reliability Grewal 2012. Uncurated online resources frequently omit conservative management options, exaggerate procedural benefits, and minimise complication risks, which distorts shared decision-making.

Clinical services curate information pathways by correcting inaccurate risk statements and providing validated materials matched to the patient's specific anatomy, comorbidity, and treatment goals. For conditions such as carotid stenosis, aneurysm repair, and limb revascularization, curated information must explicitly state the diagnosis, natural history, medical therapy, procedural options, surveillance requirements, warning symptoms, and the consequences of no intervention.

72.5Global vascular capacity planning

Essential vascular care in lower-resource settings relies on tiered capacity development rather than isolated technological procurement Stewart 2016. Purchasing advanced imaging or endovascular devices fails to improve outcomes if transport, medical supply chains, or complication-management systems are absent.

Sequential tiers of vascular capacity development:

• Prevention and risk reduction: smoking cessation, hypertension and diabetes control, antiplatelet therapy, and statin access.
• Diagnostic assessment: clinical examination and reliable duplex ultrasound, with cross-sectional imaging where feasible.
• Interventional capability: reliable medical therapy and basic operative capability, progressing to endovascular services.
• Infrastructure: supply chains for drugs, grafts, and devices, accompanied by referral systems, transport networks, and structured follow-up.
• Workforce: training, recruitment, and retention of vascular surgeons and allied specialists to sustain the tiers above.

System planning begins by mapping patient flow and local disease burden to identify the missing foundational tier, establishing basic referral links and stable medical therapies before promising advanced intervention access Stewart 2016.

72.6Areas of controversy

The clinical utility of autonomous operative artificial intelligence and computer-vision scene understanding is unsettled; current literature largely represents horizon scanning rather than proven outcome benefits, requiring prospective external validation before altering practice Liyanage 2026 · Abara 2025 · Kiang 2024.

The balance between telemedicine efficiency and the exacerbation of healthcare disparities remains contested, as digital pathways may systematically disadvantage vulnerable populations if strict equity monitoring is absent Yew 2023 · Montano 2022 · Lareyre 2022.

Jurisdictional variations in classifying clinical artificial intelligence software as a medical device create uneven regulatory thresholds for deployment, oversight, and lifecycle monitoring, complicating the adoption of global standards WHO 2023 · FDA 2021.

Vascular robotics: operator assistance versus autonomy

Robotic assistance is not synonymous with autonomous AI. Surgeon-controlled laparoscopic assistance supports sutured vascular reconstruction; catheter robots address a different task, manipulation of endovascular devices. Historical experience includes a single-centre series of 310 robot-assisted aortic and non-aortic operations: 298 were completed robotically and 10 required conversion. This was a feasibility series, not a comparison proving superiority over open or endovascular repair Novel Endovascular 2023 · Robot assisted 2016.

An endovascular first-in-human study treated five patients with lower-extremity PAD and reported robotic device manipulation without manual conversion in all five. Console-operator radiation exposure was 97.6% lower than the measurement at the procedure table; this was not a comparison demonstrating reduced patient dose or protection of the whole team. Device changes still required human assistance. This small, uncontrolled experience does not establish comparative safety, durable clinical benefit or cost-effectiveness Novel Endovascular 2023.

A 2026 synthesis of 40 studies across vascular territories, including three randomised trials, found similar comparative clinical success, major adverse cardiovascular/cerebrovascular events and mortality. Operator radiation was lower, but total procedure and fluoroscopy times were longer; patient radiation and contrast use were similar. These pooled findings do not establish superiority for lower-extremity PAD specifically Robot-assisted versus 2026.

As a service-planning implication, evaluation should specify supported devices and tasks, bedside staffing, manual rescue arrangements, operator training, and measurement of clinical outcomes, radiation and resource use. Technical success with operator-controlled robotics is not evidence of autonomous surgical capability.

Further reading

• Machine learning in vascular surgery: a systematic review and critical appraisal. (2022). Critical appraisal of vascular machine-learning evidence. Li 2022
• Computer vision applications in vascular surgery: a systematic review and critical appraisal. (2026). Computer vision in vascular surgery. Liyanage 2026