Impact of artificial intelligence on acquisition efficiency, image quality and resolution in knee MRI: a pre-post quasi-experimental study protocol

L. Guardamagna,^1,2 D. Savioni,³ L. Spagnolo,⁴ S. Haoufadi,¹ G. Marzagalli,^2,5 G. Barra,^2,6 M. Provenzano,^2,7 D. Di Napoli⁸ | ¹Department of Orthopedics and Traumatology, Clinical Institutes of Pavia and Vigevano, Pavia; ²Laboratory for Rehabilitation and Orthopedic Surgery, Institute of Care “Città di Pavia”, Department of Clinical Surgical, Diagnostic and Pediatric Sciences, University of Pavia; ³Senior Radiology Technician, Radiology Unit, Institute of Care “Città di Pavia”; ⁴Chief Radiology Technician, Radiology Unit, Institute of Care “Città di Pavia”; ⁵Education Area, Medical Direction, Clinical Institutes of Pavia and Vigevano, Pavia; ⁶Quality and Accreditation Area, Medical Direction, Clinical Institutes of Pavia and Vigevano, Pavia; ⁷Directorate of Nursing and Allied Health Professions, Clinical Institutes of Pavia and Vigevano, Pavia; ⁸Medical Director, Clinical Institutes of Pavia and Vigevano, Pavia, Italy

This single-centre quasi-experimental pre-post study evaluates the impact of artificial intelligence (AI) on knee magnetic resonance imaging (MRI) examinations. The objectives include assessing changes in image acquisition time per exam, evaluating overall image quality, and analysing image resolution. These outcomes will be measured in a randomized sample of 100 patients who underwent knee MRI, comparing 50 standard examinations (from January to June 2024) with 50 AI-powered examinations (from January to June 2026). For each period, examinations will be assigned a unique study ID, and a computer-generated random sequence will be used to select 50 cases per cohort. If a selected case meets exclusion criteria (incomplete exam, major motion artefact), it will be replaced by the next randomly selected ID. Image quality will be analysed in terms of variations in the signal-to-noise ratio (SNR). SNR assessment will be conducted on regions of interest selected from five images per patient in both cohorts, yielding a total of 250 per group. The study is expected to demonstrate reduced image acquisition times, non-inferior image quality, and improved resolution, supporting the role of AI-enhanced workflows in routine musculoskeletal MRI practice.