Clinical evaluation of a novel artificial intelligence-based image processing algorithm for two-dimensional angiography.
Authors
Affiliations (20)
Affiliations (20)
- Department of Radiology and Biomedical Imaging, Section of Interventional Radiology, Yale School of Medicine, New Haven, CT, USA. [email protected].
- Department of Internal Medicine, Section of Medical Oncology, Yale School of Medicine, New Haven, CT, USA. [email protected].
- Department of Surgery, Section of Surgical Oncology, Yale School of Medicine, New Haven, CT, USA. [email protected].
- Piedmont Heart Institute, Atlanta, GA, USA.
- Pediatric Cardiology, Arkansas Children's Hospital and University of Arkansas for Medical Sciences, Little Rock, AR, USA.
- Department of Pediatrics, Section of Pediatric Cardiology, Yale School of Medicine, New Haven, CT, USA.
- Neuroradiology and Neurological Surgery, UT Southwestern Medical Center and, Parkland Health, Dallas, TX, USA.
- Neurointerventional Surgery, Swedish Medical Center, Denver, CO, USA.
- ReVaMP Heart and Vascular Institute, Philadelphia, PA, USA.
- Department of Radiology, Division of Vascular and Interventional Radiology, Medical College of Wisconsin, Milwaukee, WI, USA.
- Siemens Medical Solutions USA, Inc, Malvern, PA, USA.
- Iowa Radiology and Department of Radiology, UnityPoint Health, Des Moines, IA, USA.
- Northwell Cardiovascular Institute, New Hyde Park, NY, USA.
- North Shore University Hospital, Manhasset, NY, USA.
- Quantitative Health Sciences, Cleveland Clinic, Cleveland, OH, USA.
- Sorin Medical, New York, P.C.NY, USA.
- Department of Cardiology, Boston Children's Hospital, Boston, MA, USA.
- Goodman Campbell Brain & Spine, Indianapolis, IN, USA.
- Center for Medical Device Evaluation, Smith Institute, Department of Medicine, Beth Israel Deaconess Medical Center, MA, Boston, USA.
- Siemens Healthineers, Forchheim, Germany.
Abstract
To evaluate the clinical performance of a novel artificial intelligence (AI)-based image processing algorithm for two-dimensional angiographic imaging compared with its predecessor across multiple imaging modes, anatomic regions, and endovascular specialties. This retrospective, comparative, randomized, paired, and blinded multicenter reader study included anonymized images acquired during clinically indicated endovascular interventions at 11 institutions in the United States and Europe. A total of 553 image series from 182 patients representing 51 procedure types across interventional radiology, neurointervention, cardiology, and pediatric cardiology were processed using both the AI-based and predecessor image processing algorithms. Thirteen board-certified physicians independently evaluated paired images using five-point Likert scales for overall image quality, contrast of clinically relevant details, noise reduction, and confidence in assessing device and vessel details. The primary endpoint was non-inferiority of overall image quality. Secondary endpoints evaluated superiority across image quality metrics using a predefined hierarchical testing strategy. The AI-based image processing algorithm demonstrated non-inferior overall image quality in 98.9% of all evaluations (95% confidence interval [CI], 98%-100%), meeting the primary endpoint across all imaging modes, anatomic regions, and predefined subgroups. Superiority of the AI-based algorithm was demonstrated for overall image quality, contrast of clinically relevant details, noise reduction, and confidence in assessing device and vessel details across fluoroscopy, roadmap, overlay reference, and native acquisition imaging. Overall image quality superiority was not demonstrated for digital subtraction angiography, likely reflecting the intrinsically higher signal-to-noise characteristics of these acquisitions. Inter-reader agreement analysis demonstrated identical ratings in 49.3% of evaluations, with 94.7% of paired ratings differing by no more than one Likert category. In this multicenter blinded reader study, AI-based image processing demonstrated non-inferior overall image quality compared with its predecessor across all evaluated imaging modes and patient subgroups, with superior performance for multiple image quality metrics in fluoroscopy, roadmap, overlay reference, and native acquisition imaging. These findings support the clinical utility of AI-based image processing for endovascular imaging and provide a foundation for future studies evaluating its impact on radiation dose and procedural outcomes.