Modern medicine has been transformed by artificial intelligence, particularly in the field of diagnostics. Such breakthroughs are doubly disadvantaged when it comes to patent prosecution and enforcement in the US, however, in an ongoing challenge for technology transfer officials.
Artificial intelligence is being successfully harnessed to improve patient outcomes, spotting disease and degenerative disorders earlier than ever before.Research universities, government labs and publicly funded health systems have often led the effort.
Examples stretch beyond well-known genomic editing tools like CRISPR-Cas9, a collaborative effort by MIT, Harvard and several affiliated hospitals capable of identifying cancer mutations for targeted treatment, among myriad other applications.
In the 1980s, University of Iowa retina specialist Dr. Michael Abràmoff began experimenting with machine learning algorithms and neural networks to identify degenerative conditions before symptoms appear. Today, automated retinal image analysis is used to diagnose diabetic retinopathy, which causes irreversible blindness when not discovered in time.
Abràmof’s IDx-DR device, rebranded in 2023 as LumineticsCore, can spot the disease in about 20 seconds and was the first autonomous AI diagnostic system to receive US Food and Drug Administration approval for use in clinical settings.
The Mayo Clinic developed an AI model capable of detecting atrial fibrillation (AFib) and assessing heart failure risk in an otherwise normal electrocardiogram (ECG).
It recently partnered with AI health technology company Anumana to commercialize various AI-ECG algorithms, including one which identifies AFib patients who stand to benefit from left atrial appendage occlusion in place of conventional blood thinner treatments.
Stroke triage and hemorrhage detection tools from the University of California system include an algorithm by researchers at UC San Francisco and UC Berkeley capable of assessing dozens of computed tomography scan images in one second, to Code Blue — a mobile app by undergraduate Ashmita Kumar which relies on AI and a smartphone’s camera and microphone to spot signs of a brain bleed in progress.
Such innovations, while clearly novel, can nonetheless be difficult to commercialize.
— MPEP update —
The prognosis for AI-related patents is touch and go.
The US Patent and Trademark Office in December gave advanced notice of changes to the Manual of Patent Examining Procedure (MPEP) that will incorporate Ex Parte Desjardins (see here).
In that ruling, an appeals review panel led by USPTO Director John Squires vacated (see here) a new, eligibility-based rejection by the Patent Trial and Appeal Board, instructing examiners to avoid evaluating AI innovations “at . . . a high level of generality.”
But the US Court of Appeals for the Federal Circuit — the destination for all patent-related appeals in the US — views the MPEP as non-binding, albeit informative, authority. It has taken a more restrictive view of AI and Section 101, ruling last year (see here) that claims which apply established methods of machine learning to a new data environment are ineligible for patenting.
— ‘Significantly more’ —
Diagnostics, meanwhile, have been on life support for over a decade.
In 2012’s Mayo Collaborative Services v. Prometheus Laboratories, the US Supreme Court said simply applying natural correlations to a diagnostic method is not enough under Section 101. Instead, the statute requires “significantly more” to the law of nature disclosed.
The following year, the justices drilled down on eligibility further when ruling, in Molecular Pathology v. Myriad Genetics, that isolated, naturally occurring DNA segments are patent-ineligible products of nature — regardless of how much experimentation was needed to achieve the isolation.
With that framework in place, the Federal Circuit in 2015 deemed a non-invasive prenatal test outside the bounds of Section 101 because it relies on conventional methods to detect natural phenomena — cell-free fetal DNA — in maternal blood.
The Supreme Court denied patent owner Sequenom’s petition for writ of certiorari in 2016, in a win for Ariosa Diagnostics.
— Risks, vulnerabilities —
Taken in the aggregate, AI-based diagnostics are at heightened risk of rejection and — for those that defy the odds and go on to be registered by the USPTO — uniquely vulnerable to legal challenge.
That could have outsized consequences for public health.
Thomas Hiesberger, partner at Dentons US LLP, said Monday* that one recent study showed that 11.1-percent of cancer, cardiovascular disease or infectious disease patients are misdiagnosed, and that 4.4-percent of those misdiagnoses resulted in serious medical harm in the form of death or permanent disability.
For technology transfer departments the current landscape hampers efforts to get a breakthrough over the finish line.
Licensing success will often hinge on whether an AI diagnostic credibly claims a technological improvement; merely reciting improvements in diagnosis accuracy or tying AI outputs to patient outcomes isn’t enough, at the USPTO or in court.
A shift in the landscape could be on the horizon, however.
In his first act as director, Squires signaled unease with the high rate of Section 101 rejections when personally signing off on two patents — one for medical diagnostics and another for distributed ledger/crypto technologies.
“These have been areas of great, but in my view unproductive, debate — too often dismissed as ‘mere business methods’ or ‘ineligible diagnostic practices.’ As President Trump recently reaffirmed Calvin Coolidge’s ageless maxim, ‘the business of America is business.’ I wanted to send a clear message with the first two patents issued on my watch: the U.S. Patent Office is open for business, especially for the technologies of tomorrow,” he said in a statement.
There is also movement in Congress.
The Patent Eligibility Restoration Act would dramatically change the risk profile and value of AI diagnostics by abolishing the judicial exceptions to subject matter eligibility articulated in Mayo, Alice and Ariosa.
While the bill in its current form does not clear a path to patentability outright — moving the battleground instead to other statutory provisions, including Sections 112 and 103 — elimination of Section 101 as a kill switch would remove the cloud of uncertainty which currently looms large over AI diagnostics.
With an estimated 370,000 deaths and 424,000 permanent disabilities flowing from medical misdiagnosis annually, per the study cited by Hiesberger, time could be of the essence.
*AUTM Annual Meeting, Feb. 9, 2026, Seattle.
Please email editors@mlex.com to contact the editorial staff regarding this story, or to submit the names of lawyers and advisers.