SIR 2024
Arterial Interventions and Peripheral Arterial Disease (PAD)
Jung H. Yun, MD
IR/DR Resident
Jefferson Einstein Hospital
Financial relationships: Full list of relationships is listed on the CME information page.
Avi Sharma, MD
Director of AI & Body Radiologist
Jefferson Einstein Hospitals
Financial relationships: Full list of relationships is listed on the CME information page.
Sara Silberstein, MD, MS (she/her/hers)
Integrated IR/DR Resident Physician, PGY2
Jefferson Einstein Hospitals
Financial relationships: Full list of relationships is listed on the CME information page.
Derek Biederman, MD
Interventional Radiologist
Einstein Medical Center, Jefferson Health
Disclosure information not submitted.
Ryan K. Lee, MD
Radiology Department Chair
Einstein Medical Center, Jefferson Health
Disclosure information not submitted.
Implementation of an AI-based workflow utilizing a commercial solution (Aidoc, Tel Aviv) for patients with acute PE involved two steps: 1) an AI-driven worklist triage system (Aidoc-PE-only) was adopted to identify the presence of PE in all computed tomography pulmonary angiograms (CTPAs); 2) an AI-driven alert system (Aidoc-PERT), integrated with a mobile workflow, sent notifications for cases with large central PE and a right ventricular to left ventricular (RV/LV) ratio >1. The study was split into three periods: pre-Aidoc (01/2019-06/2019), Aidoc-PE-only (01/2020-06/2020), and Aidoc-PERT (01/2023-06/2023). Interventional radiology reports were used as the reference standard to identify advanced interventions initiated by the Pulmonary Embolism Response Team (PERT). The report identification was facilitated by a Natural Language Processing (NLP) engine.
Results: A total of 6,111 patients who underwent CTPAs were analyzed by the AI solution, of which 71 patients had PERT interventions. The monthly mean volume of CTPAs were: pre-Aidoc, 337 scans; Aidoc-PE-Only, 316 scans; and Aidoc-PERT, 365 scans. The prevalence of PERT intervention was: pre-Aidoc, 0.84% (17/2,022); Aidoc-PE-Only, 1.16% (22/1,898); and Aidoc-PERT, 1.46% (32/2,191 CTPAs). During the Aidoc-PERT phase, 84% (27/32) of the PERT interventions were preceded by an initial CTPA scan. The system alerted on 89% (24/27) of the patients. 16% (5/32) of the PERT interventions did not have initial CTPAs; 60% (3/5) were CT CAP (Chest-Abdomen-Pelvis) aortic dissection protocol and 40% (2/5) were routine venous-phase CT CAP exams.
Conclusion:
The phased implementation of an AI-based workflow for acute PE led to an increase in clinically appropriate advanced interventions. The Aidoc-PERT period showed the highest prevalence of advanced interventions. Notably, 16% of patients undergoing PERT interventions were identified on non-CTPA exams, underscoring the necessity for AI-driven workflows to include incidental PE, which have since been implemented. These findings suggest that integrating AI into acute PE diagnostic and interventional workflows can enhance quality and timeliness of patient care. Further analysis aims to include clinical outcomes such as LOS and M&M.