Observational evaluation of AI-assisted measurements and reporting for enhanced workflow efficiency in leg and foot radiographs.
Authors
Affiliations (7)
Affiliations (7)
- Department of Musculoskeletal Imaging, Cochin Hospital, AP-HP, Paris, France. Electronic address: [email protected].
- Gleamer, Paris, France.
- Réseau d'Imagerie Sud Francilien, Lieusaint, France.
- Department of Musculoskeletal Imaging, Cochin Hospital, AP-HP, Paris, France.
- Department of Radiology, VA Boston Healthcare System, Boston University School of Medicine, Boston, MA, USA.
- Gleamer, Paris, France; Réseau d'Imagerie Sud Francilien, Lieusaint, France; Ramsay Santé, Clinique du Mousseau, Evry, France.
- Gleamer, Paris, France; Réseau d'Imagerie Sud Francilien, Lieusaint, France.
Abstract
Radiographic musculoskeletal (MSK) measurements are essential for diagnosis and surgical planning, but they remain time-consuming and prone to variability. Artificial intelligence (AI) can address these limitations by automating both measurements and reporting. This study assessed the impact of AI-assisted measurements and reporting on workflow efficiency for leg and foot radiographs using BoneMetrics and AutoReport (Gleamer, Paris, France). Leg and foot radiographs were collected retrospectively. The ground truth was established by a senior MSK radiologist through manual annotation of key measurements. The same measurements were performed by a junior radiologist first manually, and subsequently with AI assistance. Reports were then generated by the radiologist via voice dictation and subsequently using AI-generated reports. These sessions formed the basis for simulated workflows: manual measurements with dictation, AI-assisted measurements with dictation, AI-assisted measurements with automated reporting.Measurement accuracy was compared between the AI solution and the junior radiologist. A total of 98 leg radiographs and 101 foot radiographs were analyzed by the junior radiologist. Measurement time was significantly reduced from 166 to 40 s (p < 0.001) and reporting time from 80 s to 33 s (p < 0.001) with AI assistance. When combining both steps into simulated workflows, total interpretation time decreased from 246 s in the fully manual workflow to 73 s in the fully automated workflow, representing a 70 % gain in efficiency. The AI solution demonstrated high measurement accuracy with performance comparable to that of the junior radiologist. This study demonstrated that AI assistance improved workflow efficiency in leg and foot radiography without compromising measurement accuracy. Integrating automated measurements and reporting in the end-to-end workflow holds promise for streamlining clinical practice.