LOS ANGELES, October 6, 2026
GigHz, a physician-founded software and research company, has highlighted findings from a peer-reviewed study led by its founder, Pouyan Golshani, MD, showing that 76.5% of 1,430 FDA authorization records for artificial intelligence and machine learning-enabled medical devices were reviewed under the Radiology panel. Published in Cureus, the analysis examines three decades of FDA authorization patterns and raises questions about the infrastructure, data accessibility, and clinical workflows needed to expand AI-enabled medical devices across other areas of healthcare.
Radiology Dominates FDA AI Device Authorizations
The study by Golshani and Mary S. Joseph analyzed entries in the FDA’s public AI-enabled medical device list with authorization dates ranging from September 1995 through December 2025. Of the 1,430 authorization records reviewed, 1,094 were associated with the Radiology review panel. Radiology, Cardiovascular, and Neurology together accounted for 90.6% of all records, demonstrating a strong concentration of FDA-authorized AI and machine learning-enabled medical devices within a relatively small number of clinical areas. The researchers reported that 331 AI device authorizations occurred in 2025, reflecting the rapid growth of the field. Annual authorizations averaged approximately 1.8 per year between 1995 and 2014, compared with an average of 264 per year from 2023 through 2025. The study also found significant concentration among developers. Of 740 companies represented in the dataset, 502, or 67.8%, had a single authorized AI device, while 13 companies, representing 1.8% of companies, accounted for 247 devices, or 17.3% of all records.
Digital Infrastructure May Support AI Adoption
According to the study, the concentration of AI medical device authorizations in radiology may be partly associated with the specialty’s established digital infrastructure. Radiology has long relied on digital medical images, standardized file formats, and systems that transmit imaging studies to clinicians for interpretation. These characteristics can provide developers with structured data environments in which AI technologies can be integrated into clinical workflows. However, the researchers emphasize that the authorization pattern does not establish why radiology has attracted a larger share of AI medical devices, nor does it mean radiology is easier to automate. The analysis is descriptive and focuses on FDA authorization records rather than determining causal factors behind the distribution. Outside the leading review panels, the study identified substantially fewer authorization records. Pathology accounted for nine records, Microbiology for six, and Obstetrics and Gynecology for four across the study period. No authorizations were recorded under a Psychiatry or Behavioral Health review panel. The researchers caution that FDA review-panel categories do not directly correspond to every specialty or clinical setting where a medical device may ultimately be used.
Advertisement
Study Highlights Clinical AI Delivery Challenges
The findings also point to a broader challenge for clinical AI development: moving from available data to useful clinical decisions. GigHz founder Golshani noted that many guideline-based decisions in areas such as internal medicine could potentially benefit from improved decision support, but relevant information may be distributed across clinical notes, laboratory results, medications, and previous visits. The study therefore highlights the importance of defining a specific clinical task, ensuring necessary data are accessible, and testing AI tools within the workflow where they will ultimately be used. Different applications, such as documentation systems and treatment recommendation tools, may require different levels of evidence, validation, and safeguards. Importantly, the analysis measures FDA authorization records rather than clinical adoption, patient benefit, or physician replacement. The FDA’s AI-enabled device list is also not comprehensive and does not capture the full range of healthcare AI, including software functions that fall outside medical-device regulation. The study provides a descriptive view of authorization trends while highlighting the continuing challenge of bringing effective AI technologies into broader areas of healthcare.
Source: GigHz press release
