Clinical Scorecard: AI Applications in Radiology: An Overview of 173 Market-Available Solutions and Their Supporting Evidence
At a Glance
Category
Detail
Condition
Radiological imaging and diagnostics
Key Mechanisms
Machine learning and deep learning-based AI tools for imaging analysis
Target Population
Patients undergoing radiological imaging across various organ-based subspecialties
Care Setting
Radiology departments and imaging centers within European healthcare systems
Key Highlights
173 CE-certified radiological AI products from 90 vendors were evaluated, showing market growth with some recent decline in new product introductions.
Only a minority of AI products have peer-reviewed clinical evidence, with most studies focusing on technical performance rather than patient outcomes or socio-economic impact.
Regulatory requirements (MDR) mandate clinical evidence and prospective monitoring for CE certification, but publicly available data remain limited.
Guideline-Based Recommendations
Diagnosis
Use peer-reviewed evidence to assess AI product diagnostic accuracy and technical performance before adoption.
Consider hierarchical efficacy levels from technical validation to patient outcomes when evaluating AI tools.
Management
Integrate AI tools cautiously into existing radiology workflows, addressing data privacy and interoperability challenges.
Prioritize AI products with demonstrated clinical relevance and multicenter, independent validation.
Monitoring & Follow-up
Conduct prospective monitoring of AI product performance post-implementation as required by MDR.
Encourage multicenter and multinational data collection to ensure generalizability and robustness.
Risks
Be aware of limited evidence on AI impact on clinical decision-making and patient care.
Recognize potential barriers including unclear return on investment and integration difficulties.
Patient & Prescribing Data
Patients undergoing diagnostic radiological imaging across multiple organ systems
Current AI products primarily support diagnostic accuracy; evidence for improved clinical outcomes and socio-economic benefits is limited and evolving.
Clinical Best Practices
Evaluate AI products using a hierarchical model of efficacy covering technical, diagnostic, clinical, and socio-economic levels.
Prefer AI solutions with peer-reviewed, vendor-independent studies involving diverse, multicenter datasets.
Maintain awareness of regulatory requirements including clinical evidence submission and ongoing performance monitoring.
Address workflow integration and data privacy proactively to facilitate adoption.
Monitor emerging evidence to guide evidence-based implementation and update clinical protocols accordingly.
Joint clinical consensus outlines evaluation and management considerations for arrhythmias, coronary atherosclerosis, aortic dilatation, myocardial fibrosis, and related findings in older competitive athletes.
