The human brain is a biological marvel with an enormous amount of raw processing power. Despite its enormous potential, its innate inefficiencies mean that, for many tasks, it doesn’t stand a chance compared to a specifically trained artificial intelligence (AI) and machine learning (ML) models. It takes years of dedicated study to train a human brain in the intricacies of medical diagnosis: spotting patterns, considering influencers, and drawing a conclusion. For example, a great volume of experience is needed to be able to look at an image and determine whether a patient is suffering from a particular disease; whether the image is a chest x-ray, a retinal scan, or an endoscopy. To a patient seeking treatment, this could mean having to wait weeks before any expert is able to see them. An AI, however, can be trained on thousands of images in significantly less time than it takes to complete medical school and could be used to increase both the efficiency and accuracy of this process. What does this mean in practice - are we close to effectively outsourcing human intelligence?
A high-profile example of AI used in diagnostics is the collaboration between Google Health and Moorfields Eye Hospital in London. Together, they’ve been analysing thousands of retinal scans to diagnose acute macular degeneration (AMD), amongst other diseases. It is hoped the project will lead to more robust predictions of AMD development and the ability to catch the disease at an earlier stage. In other fields of medicine, such as oncology, systems using AI are reporting similar levels of diagnostic accuracy to trained physicians (i.e., humans).
AI is on the cusp of becoming commonplace in digital healthcare. The benefits could be tremendous. Increased sensitivity and accuracy of diagnosis will have a positive impact on patient welfare and will ease time pressure put on healthcare employees. Protecting innovation and invention in this cross-sector arena will be crucial to keeping up with the speed of evolution in this new area.
However, securing a patent for an AI-related invention is not always straightforward. You can read more about the challenges of AI patent drafting in our previous article, ‘AI Meets IP: The patent challenge’. Article 52 of the European Patent Convention (EPC) specifically excludes “programs for computers” and “mathematical methods” from patent eligibility “as such”. This can present challenges. Developing a new, improved algorithm might feel inventive, but drafting a claim for the algorithm alone is likely to face Examiner’s objections.
The key to gaining protection for such an invention is tying the claims of the AI, or AI-implemented method, to its technical effect or advantage for the task in question. The European Patent Office (EPO) treats AI and ML in a similar way to mathematical methods. In other words, AI or ML that don’t serve a particular technical purpose will be found to lack a technical effect, and therefore will not be considered to involve an “inventive step”. To overcome this problem, AI or ML that has a clear technical effect, such as improving the way the computer is able to complete a specific task or providing an effect that occurs outside of the computer, will find more favour at the European Patent Office (EPO). One such area that provides a good starting point for showing a technical effect is disease diagnosis.
However, this coast isn’t clear just yet. Overcoming computer program-related exclusions by using the diagnostic quality of the invention may bring another exclusion into play. European patent law is set up to avoid physicians being impeded by patent rights, understandably. Article 53 of the EPC states that patents shall not be granted for “diagnostic methods practised on the human or animal body”. Experience in this area has shown that strategies are available to overcome this exclusion, and AI/ML (or software-implemented) inventions in this field can benefit from a unique exception to the way Article 53 is interpreted.
In order to be patentable, a diagnostic method must not be “practised on the human or animal body”. In one of the key decisions in this field, G1/04, the Enlarged Board of Appeal (EBA) stated that if the performance of the invention “implies any interaction with the … body, necessitating [its] presence” it would be ineligible for patent protection. However, in the same decision the EBA stated:
“However, if … some or all of the method steps of a technical nature … are carried out by a device without implying any interaction with the human or animal body, for instance by using a specific software program, these steps may not be considered to satisfy the criterion “practised on the human or animal body”, because their performance does not necessitate the presence of the latter”
In this decision the EBA appeared to indicate that any technical step of a diagnostic procedure implemented by software would not be considered to interact with the body.
A patent application in the world of digital healthcare tested the boundaries of this interpretation. The application in question concerned a heart monitor which monitored electrical signals to identify arrythmia. The monitor was both wearable and portable, such that it could take continuous readings without requiring hospitalisation, and made use of a Kohonen neural network in order to process electrocardiographic signals. The signal data were collected, stored and compared to stored values. The monitor could then give an indication of whether the heart was behaving arrhythmically – in essence, a diagnosis. Crucially, as the monitor was portable and intended to be attached to the patient, it removed the need for continuous monitoring by a trained physician.
In decision T 0598/07, the Examining Division found that the invention did carry out diagnostic steps. While this would usually lead to an objection under Article 53 EPC, the Examining Division found that these steps were “being performed by a computer, they are not performed on the human or animal body”. Thus, it appears that the exemption, characterised in the earlier G1/04 decision, extends even to computer programs carrying out diagnostic steps in wearable devices. In other words, the invention may be attached to the body, but it is not being performed on the body!
This example shows us at least one definitive way in which AI-implemented diagnostic inventions may be patentable in Europe. We believe that inventions focussing on the union of AI with digital healthcare will certainly continue to drive medical progress, given the significant benefits to both the patient and healthcare system. AI may not have wholly replaced human brain power yet – but it’s clearly on the pulse of European MedTech innovation.
For any questions or comments on this material, please contact Sarah Lau, Matt Aldridge, Michael Newton, Harry Harden or your usual Kilburn & Strode advisor.