The FDA recently announced that more than 520 FDA-approved artificial intelligence (AI) medical algorithms are now available in the United States. AI algorithms in medicine are not new — the FDA approved the first AI algorithm in 1995. Growth was slow over the next 18 years. AI usage has dramatically risen in the last decade, with over 300 AI algorithms approved from 2019 through 2022. In October 2022, the FDA approved 178 artificial intelligence (AI) and machine learning (ML) applications.
AI now exceeds human performance in many medical areas, especially medical imaging. Medical device manufacturers are incorporating AI and ML applications to increase diagnostic accuracy and enhance patient care.
AI in Medical Imaging
One of AI’s strengths is the ability to distinguish patterns within data. Not surprisingly, medical imaging is the primary beneficiary of AI’s image recognition power.
75% of all AI medical applications are in medical imaging.
Most medical imaging applications appear in subspecialties — by utilizing thousands, or even tens of thousands, of images, AI learns patterns of the varied presentations of specific diseases, including certain cancers, breast density, strokes, cardiac abnormalities, and brain abnormalities.
Even in other medical fields, AI applications incorporate imaging, including cardiology, dental, gastroenterology, obstetrics/gynecology, ophthalmology, pathology, surgery, and urology.
Keith J. Dreyer, DO, PhD, FACR, American College of Radiology (ACR) Data Science Institute chief science officer, offers more insights on AI in medical imaging. Dreyer serves as vice chairman of radiology at Massachusetts General Hospital and chief data science and information officer for both Massachusetts General Hospital and Brigham and Women’s Hospital radiology departments. He explains that AI is not just for diagnoses and can be used in various ways.
- Automatically identify life-threatening findings even before a radiologist views the images
- Automation of reports, including contouring, quantification, and auto-complete text
- Provide clinical support for subsequent steps in patient care
- Guidance AI to assist imaging technologists in achieving superior images, regardless of users with little experience with the equipment or the anatomy
- Automated identification and contouring of specific types of tissue and organs
- Enhanced image reconstruction to increase image resolution
- Increase workflows
Non-clinical Use of AI in Healthcare
FDA approval isn’t required if AI applications do not influence/drive clinical patient care. Healthcare facilities of all kinds are taking advantage of new AI solutions to fill in the gaps caused by staffing shortages and perform extremely labor-intensive tasks.
At the last Healthcare Information and Management Systems Society annual meeting, more than 1400 vendors showcased their clinical and non-clinical AI healthcare applications. Non-clinical AI can be used in a wide variety of healthcare tasks.
- Revenue cycling management efficiency
- Health-tracking apps
- Detection and solutions for gaps in health equity
- Facility tracking for the length of stay, readmissions, bed turnover statistics, and early sepsis detection
- Improving patient care, preventive care, and timely screenings
The FDA continues to work with AI creators because some newly-launched AI algorithms continue to learn after deployment. After an AI solution is deployed into live healthcare environments, questions remain on oversight and monitoring.