Five Findings from the FDA’s List of AI Medical Devices

An increasing number of medical devices are incorporating artificial intelligence or machine learning (AI/ML) functionalities. The FDA updated its list of AI/ML-enabled devices on May 13, 2024, adding 191 newly approved devices, bringing the total to 882.

The most recent ten approvals are:

1. Eko Health’s Low EF AI: Detects key heart failure indicator Low Ejection Fraction (EF) within 15 seconds during a standard physical exam.
2. Coreline Soft’s Aview CAC: A computer-aided detection (CAD) software designed to help radiologists detect lung nodules (3-20 mm in diameter) during chest CT scans in asymptomatic individuals. The algorithm has been validated using non-contrast CT images, most of which were acquired on Siemens SOMATOM CT series scanners; therefore, it is recommended to limit the device’s use to the Siemens SOMATOM CT series.
3. Siemens’ NAEOTOM Alpha: Utilizes quantum technology, making it the world’s first photon-counting CT, an upgrade in computed tomography. Its detectors convert signals directly, offering high-resolution images with minimal doses, spectral information with each scan, and better contrast at lower noise levels.
4. Siemens’ MAGNETOM Terra: The first 7T MRI scanner for diagnostic imaging. Its unique dual-mode allows patients to switch between clinical and research operations and uses separate databases to distinguish between clinical and research scans. Ultra-high field (UHF) technology represents the latest upgrade in MRI.
5. Siemens’ syngo.CT DE Brain Hemorrhage: Helps healthcare professionals visualize iodine concentration and distribution in the brain. It can compute virtual non-contrast images from enhanced scans, avoiding a second scan.
6. United Imaging’s uMR Omega: A 75 cm ultra-wide bore 3T MRI with a 60 cm field of view and ultra-high uniformity (typical value of 0.029 ppm @ 30 DSV), achieving perfect skin fat saturation and consistent image quality.
7. Edwards Lifesciences’ Algorithm: Helps clinicians predict the risk of systemic hypoperfusion events in patients undergoing advanced hemodynamic monitoring.
8. Anderson Cancer Center’s Radiation Planning Assistant: Assists clinicians in predicting the risk of systemic hypoperfusion events, aiding in cancer treatment planning.
9. Brainlab’s Automatic Registration: A spinal navigation software. Image registration keeps navigation updated with anatomical changes while minimizing radiation exposure to patients and the surgical team.
10. Avicenna.AI’s CINA-ASPECTS: Detects suspected intracranial hemorrhage on non-contrast CT scans, reducing turnaround time for patients with head trauma and hemorrhagic stroke.

Below are five findings from the list of devices:

1. Submission Slowed During Covid-19 but Resurged in the Last Two Years

Since 2015, the number of AI/ML device submissions to the FDA has increased annually. The largest growth occurred between 2019 and 2020, with a decrease in submissions from 2020 to 2022. However, the following two years saw an increase of over 170 submissions each year.

2. Siemens and Aidoc Have the Most Approved AI/ML Devices, With China’s United Imaging Holding 16 Licenses

Among these companies, Aidoc, less familiar to Chinese readers, is headquartered in Tel Aviv and produces triage systems for radiologists. Brainlab, headquartered in Munich, collaborates with ZimVie to develop surgical navigation for spinal surgeries.

3. Radiology Dominates AI/ML Device Approvals

Majority of AI/ML devices authorized to date belong to the field of radiology, accounting for 76% of approvals. Radiological imaging data was digitized earlier compared to other applications. The 21st Century Cures Act also mandates that any device analyzing medical images must be regulated as a medical device. In 2022, the FDA sought to clarify through final guidelines which types of software should be regulated as devices, such as those using electronic health records to flag potential sepsis cases.

4. None of the Listed Devices Use Generative AI

Despite growing interest in large language models like ChatGPT, such technologies have not been applied to any approved devices. The FDA states that it has not yet approved any device using generative AI or general AI driven by large language models. The complexity of models in the list varies. Some are “shallow,” with fewer than two hidden layers between inputs and outputs, while others use deep learning. Typically, models use a combination of methods to achieve results, such as using one model to generate features and another for classification. Examples of approved devices this year include an algorithm developed by Edwards Lifesciences to help clinicians predict the risk of systemic hypoperfusion events in patients undergoing advanced hemodynamic monitoring, and a radiotherapy planning assistant developed by the Anderson Cancer Center to help clinicians predict the risk of systemic hypoperfusion events, aiding in cancer treatment planning.

5. Most Devices Approved via the 510(k) Pathway

Vast majority of devices in the list were approved through the 510(k) pathway, with only 12 devices receiving De Novo clearance and four approved through the PMA channel.

About Accel

Accel Groups is a dedicated team of seasoned professionals in regulatory, clinical and market access. We help medical device and IVD manufacturers with complete product life cycle solutions from preclinical, strategy, clinical evaluation and trial, regulatory submission, to post market surveillance. Our regional principals have a minimum of 10+ years’ experience in their specialty to enable us to provide the regional in-depth expertise but with global one-stop shop solutions. Ask us about our Start-up kits for start-up companies with key countries regulatory pathway.