Software As A Medical Device: Can It Be Patented?

In general, if computer software meets the definition, software can be deemed a “medical device” by the US Food and Drug Administration (FDA) and can also be patented, though patenting software can be difficult. The standards used for patentability and what constitutes a “medical device” for the FDA are, however, different.

FDA Definition of SaMD

The FDA uses the definition developed by the International Medical Device Regulators Forum (IMDRF), a group of medical device regulators from around the world: SaMD is “software intended to be used for one or more medical purposes that perform these purposes without being part of a hardware medical device.” The FDA chaired the IMDRF working group, formed in 2013, that wrote this definition, and it still appears on the FDA’s Software as a Medical Device page. On December 8, 2017, the FDA also adopted the IMDRF’s clinical evaluation principles for SaMD as guidance, but it withdrew that guidance on January 6, 2026.

Under this definition, SaMD is software that provides an “independent review” or provides “significant” information in treatment or diagnosis of a medical or health condition. Software that merely retrieves information, optimizes performance of a medical device or that organizes data is not considered SaMD. Likewise, software that is incorporated into and serves to drive particular hardware for a medical device is not considered SaMD.

SaMD is expected to provide information that is “significant” to making a healthcare decision. Under the IMDRF framework, which the FDA summarizes on its Global Approach to Software as a Medical Device page, the information falls into one of three levels:

• Treat or diagnose: the information is used to take an immediate or near-term action, such as screening for or detecting a disease or condition.
• Drive clinical management: the information aids treatment or diagnosis, or triages or identifies early signs of a disease to guide the next test or treatment.
• Inform clinical management: the information does not trigger an immediate action, for example by setting out treatment options or pulling together relevant clinical information.

The framework also weighs how serious the medical condition is: critical, serious, or non-serious. Software that treats or diagnoses a critical condition falls in the highest-impact category (IV), while software that only informs management of a non-serious condition falls in the lowest (I).

In addition to providing “significant” information, the SaMD should have a level of “independence.” That is, in general, SaMD should be capable of running on any general purpose computing platform and should, at least in theory, be able to accomplish its medical task if separated from its “normal” housing.

The KardiaBand for the Apple Watch

In November 2017, the FDA cleared the first medical device accessory for the Apple Watch, AliveCor’s KardiaBand electrocardiogram reader. The KardiaBand replaced the original Apple Watch band and paired a sensor with an artificial intelligence software application that detected abnormal heart rhythm or atrial fibrillation. The app learned what was “normal” for each user, establishing a baseline, checked the user’s heart rate every five seconds, and prompted the user to take a reading if something looked abnormal. That data could then be shared with a healthcare professional. AliveCor stopped selling the KardiaBand in 2019, after Apple built electrocardiogram recording into the Apple Watch.

Taking the KardiaBand as an example, the software component of the KardiaBand should be considered SaMD if the software can perform its function on a normal computing platform and if the software can function separately from the band, even if it needs to be attached to some other sensing mechanism. The fact that SaMD might need to be interfaced with other medical hardware does not take it outside of the definitions used by the IMDRF and the FDA.

How Is AI Used in Software as a Medical Device?

Much SaMD now runs on artificial intelligence (AI) or machine learning algorithms. AI-based SaMD can provide early detection of patient illness or a treatment profile customized to a patient’s exact needs. AI can also run real-time health risk assessments or predictive simulations to prevent complications during or after a medical procedure. All of this is possible because of the underlying algorithms of the SaMD.

In April 2018, the FDA authorized IDx-DR, software that analyzes eye images to detect diabetic retinopathy, as the first device to provide a screening decision without a clinician also interpreting the image or results. Today AI runs in diagnostic tools, robotic surgical-assistance tools, and wearable health monitors that communicate with a patient’s smartphone, and the FDA’s public list of AI-enabled medical devices includes more than 1,600 devices.

Locked vs. Adaptive Algorithms in AI Medical Devices

According to the FDA, the algorithms used in medical device software range from fairly static, known as locked algorithms, to highly dynamic, known as adaptive algorithms. A locked algorithm provides the same output every time it is given a particular input; decision trees and static look-up tables are common examples. An adaptive algorithm is a continuous learning algorithm, so it can produce a different output for the same input over time as it changes its behavior based on what it has learned.

Adaptive algorithms capable of machine learning that are incorporated into new medical technologies can save time, reduce medical errors, and improve patient care. One prominent area of medical technology wherein the software as a medical device utilizes machine learning is the field of diagnostics, such as reviewing patient scans for signs of cancer, monitoring a patient’s physiological state, and administering a medication appropriately based on the observed patient data.

Does every update to an adaptive algorithm need a new FDA submission? Not if it follows a predetermined change control plan (PCCP). Under the FDA’s final PCCP guidance for AI-enabled devices (issued December 2024, updated August 2025), a manufacturer can describe planned modifications, how they will be validated, and their impact in its 510(k), De Novo, or PMA submission. A change consistent with an authorized PCCP does not require a new 510(k) or PMA supplement. A broader draft guidance on AI-enabled device lifecycle management, issued in January 2025, has not been finalized.

Software as Medical Device: Patentability

As noted, even though the FDA may consider certain software to be a medical device, that designation by the FDA will have little impact on the patentability of SaMD. Furthermore, patenting medical software, as software, is difficult but not impossible. With respect to SaMD, if the software is incorporated into a medical technology, it may be possible to obtain patent protection for the software as it pertains to the medical device.

SaMD must still meet the subject matter eligibility test of 35 U.S.C. § 101 under the Supreme Court’s Alice framework: claims directed to an abstract idea, such as analyzing data on a generic computer, are rejected, while claims that improve how a computer or other technology works can pass. Software inventions are potentially patentable when claimed as a method or process with enough specificity to clearly convey what the invention is.

Can AI in a Medical Device Be Patented?

Yes, in many cases. The device that uses AI-based SaMD can be eligible for a medical device patent, and the software processes behind its algorithms can be eligible for software patent protection. The key is to claim a specific technical improvement, not the general idea of applying AI to medical data.

In Recentive Analytics, Inc. v. Fox Corp. (2025), the Federal Circuit held that patents that “do no more than claim the application of generic machine learning to new data environments, without disclosing improvements to the machine learning models to be applied,” are ineligible. By contrast, in Ex parte Desjardins (2025), a precedential USPTO decision, claims to a method of training a machine learning model were eligible because the application explained how the training improved the way the model itself operates. The USPTO added Desjardins to its examining manual in December 2025, alongside its July 2024 eligibility guidance update for AI. So an AI medical device application should explain what the model, its training, or its link to sensors and hardware does better.

AI is now common across the patent system. A USPTO report, Inventing AI (October 2020), found that the share of all patent applications containing AI grew from 9% in 2002 to nearly 16% in 2018, and that by 2018 AI appeared in more than 42% of the technology subclasses the USPTO uses.

How to Protect AI Software IP in Medical Devices

Before an AI medical device can be sold, the unmet medical need must be investigated, the AI software programmed, debugged, tested, and integrated into the device hardware, and the product prototyped, tested, and submitted to the FDA. With many companies racing to meet the same needs, IP protection guards that investment from unauthorized copying or use.

Companies can rely on trade secret, and to some degree copyright, to protect the IP behind the AI. But trade secret protection lasts only while the information stays confidential, and copyright protects the code as written, not the underlying method. That is why it is critical to evaluate each AI or SaMD invention for patent protection and pursue it where appropriate.

Need Help Protecting the IP in Your SaMD or AI Medical Device?

A software as a medical device attorney can help you determine whether patent protection is right for you, weigh it against trade secret and copyright, and prepare a patent application for your AI or SaMD invention. The professionals at The Rapacke Law Group can be reached by calling 954-951-0154 to schedule a free initial consultation, or you can fill out this online contact form.