Clinical Report: AI in Radiology - Overview of 173 CE-Certified Market Solutions

Overview

This study systematically evaluated 173 CE-certified radiological AI products available by March 2023, analyzing their supporting peer-reviewed evidence and clinical efficacy. Findings reveal an increase in AI solutions since 2020 but highlight persistent gaps in evidence regarding clinical outcomes and real-world impact.

Background

Artificial intelligence has the potential to enhance diagnostic accuracy and optimize workflows in radiology, yet its adoption remains limited due to concerns about data privacy, integration challenges, unclear return on investment, and insufficient clinical evidence. European Medical Device Regulation mandates clinical evidence for CE certification, but such data are often not publicly accessible, making peer-reviewed literature crucial for evaluation. Prior analyses showed limited published evidence supporting many AI products, primarily focusing on technical performance rather than patient-centered outcomes. This study updates the landscape by assessing evidence evolution from 2020 to 2023.

Data Highlights

A total of 173 CE-certified radiological AI products from 90 vendors were analyzed, including 85 products present in both 2020 and 2023 datasets and 88 newly listed products since 2020. The annual introduction of new CE-certified AI products peaked at 44 in 2020, then declined to 24 in 2021 and 4 in 2022. Peer-reviewed evidence remains limited, with many studies focusing on diagnostic accuracy rather than clinical decision-making or patient outcomes.

Key Findings

• 173 CE-certified radiological AI products were identified as of March 2023, a substantial increase from 100 products in 2020.
• Only a minority of AI products have published peer-reviewed evidence, with most studies emphasizing technical and diagnostic accuracy (efficacy levels 1 and 2).
• Evidence addressing higher efficacy levels—clinical decision-making, patient outcomes, and socio-economic impact—is scarce.
• Multicenter collaborations, data diversity, and vendor-independent studies remain limited but are critical for robust validation.
• The annual number of new CE-certified AI products peaked in 2020 but declined in subsequent years, suggesting market maturation or regulatory impacts.
• Products with longer market presence tend to show more mature evidence profiles compared to newer entrants.

Clinical Implications

Clinicians should critically appraise the evidence supporting AI tools, recognizing that most currently available products lack comprehensive data on clinical impact and patient outcomes. Adoption decisions should consider the maturity of evidence, vendor independence, and applicability to local clinical workflows. Continued emphasis on multicenter, prospective studies is essential to validate AI benefits and facilitate integration into routine practice.

Conclusion

While the number of CE-certified radiological AI products has grown substantially, robust clinical evidence supporting their real-world effectiveness remains limited. Addressing evidence gaps through rigorous, patient-centered research is vital to realize AI's full potential in radiology.

References

1. van Leeuwen et al 2020 -- Evaluation of CE-certified radiological AI products
2. European Medical Device Regulation (MDR) -- Clinical evidence requirements for AI products