Patenting AI inventions in the MedTech space
Artificial intelligence and machine learning models face similar types of patentability challenges as ‘classic’ software inventions at the EPO and UKIPO. In particular, for an AI invention to be deemed patentable, the applicant will need to demonstrate that the invention solves a ‘technical’ problem in a non-obvious manner. Our in-depth study provides detailed guidance on patenting software inventions in Europe. An abstract machine learning model on its own is unlikely to be patentable.
This generally means that a machine learning model must be claimed in a technical context, with the features of the claim reflecting how the technical problem is solved.
This provides a strong opportunity for protecting AI inventions in the MedTech space, as the application of machine learning techniques in a MedTech context can provide the invention with the required technical character to meet patentability requirements.
AI is already used in a large variety of applications in the MedTech space, here are some examples:
- AI-controlled or AI-assisted devices and systems - such as robot-assisted surgical systems, AI-controlled radiotherapy, AI-controlled imaging systems, medical robots/assistants.
- AI-based health monitoring and diagnostics - such as analysis of output data from sensors and/or wearables, health recommendations based on sensor outputs.
- AI-based image analysis and interpretation - such as real-time analysis of radiography images or surgical images.
- Use of AI to assist in designing medical devices - such as designing implants or prosthetics with an AI tool.
Generally speaking, such applications of AI techniques in the MedTech space involve interactions of a machine learning model with a piece of hardware (e.g. medical equipment) and/or with signals obtained from hardware. This provides a clear technical context for use of the machine learning model, such that the invention will in principle be patentable – provided the claimed invention is non-obvious over the prior art.
The development of AI tools for use in the MedTech space can present a variety of different challenges, such as how to obtain suitable training data, optimising the training process to yield a desired level of accuracy of the machine learning model, and integration of the model with hardware and/or as part of a product. This raises the question: what aspects of AI inventions can be patented in the MedTech space? Below, we look at different ways in which MedTech AI inventions could be protected.
The trained machine learning model and its use
Often, a MedTech AI invention results in a new product, tool or service incorporating a trained machine learning model. For example, this could be in the form of a medical system which uses the trained model, or a consumer device (e.g. a wearable device) which interacts with the trained model in some way. From a strategic point of view, it can be important to try and cover the end use of the trained model, as this is typically where the greatest commercial value of the invention resides. Moreover, infringement of end use claims may be more readily detectable, whereas in practice it can be more difficult to determine if claims around the training method or training dataset are infringed.
Of course, for the trained machine learning model and its use to be patentable, it must support an inventive step (i.e. be non-obvious). This could arise in various ways, depending on the nature of the invention. For instance, the selection of a given machine learning model for tackling a particular problem may be seen as inventive, if the selection is non-obvious. As another example, the manner in which the model interacts with a device and/or analyses received data may be considered as inventive.
Depending on the invention, the trained machine learning model and its use could be claimed in a variety of ways. For example, an applicant could claim a computer-implemented method where the trained machine learning model is used to perform certain tasks. The tasks could include various interactions of the model with a physical device, for example, controlling the device in some way (e.g. in the case of AI-controlled radiotherapy), or receiving data from the device and generating an output in response to the received data. Such an invention could also be claimed as a computer system or program including the trained machine learning model which is configured for performing the tasks. Where specific hardware arrangements are used to implement the system, this could also be reflected in the claims. In general, a combination of system claims and method claims can be used to provide broad protection around the invention and capture different types of infringement.
Training of the machine learning model
In some cases, the process used for training the machine learning model can be protected, where the training process itself supports an inventive step. For example, if it is found that a particular training process is advantageous or provides a strong performance boost for a given MedTech application, then it may be possible to obtain a patent for the process of training the model. Of course, the training process would need to be tied to the technical context of the MedTech application, for example by specifying the types of training data used, and/or the task which the machine learning model is configured to perform.
Claims directed to the techniques used for training the machine learning model can provide useful additional protection around the invention. For example, this would enable coverage of a third party who is training the model according to the invention, even if they do not go on to use the model themselves. Such claims could be directed to a computer system or program configured to carry out the training process, as well as the training method itself (e.g. the steps used in the training method).
“When it comes to protecting IP in AI applications to biotech and pharma, patents are only one piece of the puzzle. The combination of patentable innovations that have unimaginable power to benefit the human condition, in a world where almost all of these use at least one open-source software, collaborations are often essential and regulatory considerations are never far off of sight makes it the most thrilling field to work in. Your very complicated puzzle is our passion to solve!”
Camille Terfve
Partner, Patent Attorney
Obtaining the Training dataset
MedTech AI inventions typically rely on the collection of large datasets for training the machine learning model. Obtaining the training datasets can represent significant investments in time and money, particularly if labelling of the datasets is required. Accordingly, where possible, it may be desirable to protect the process via which the training dataset is obtained.
For some MedTech AI tasks, highly specialised types of data may be required, such that obtaining the required data may present a technical challenge that needs to be overcome. Additionally, the manner in which training data is compiled and prepared could contribute to a technical effect, for instance in terms of performance improvements in the training process and/or the resulting machine learning model. Therefore, depending on the details of the training dataset and how it is obtained, it is possible that the process of obtaining the training dataset could be viewed as technical and supporting an inventive step. Patent claims could, for example, be directed to a method for obtaining the training dataset, and/or to a system configured to obtain the training data. On the other hand, if standard data collection and pre-processing techniques are used, then such a process is unlikely to be patentable.
Whilst the process of obtaining the training dataset may be protectable, it is likely to be much more difficult to obtain patent protection for the training dataset itself. Indeed, the exclusion from patentability around ‘presentations of information’ at the EPO and UKIPO would likely come into play. For the dataset itself to be patentable, one would likely need to show that a technical effect derives from the structure of the data and the manner in which it interacts with the machine learning model, regardless of the actual content of the data. However, indirect protection for the dataset may in some cases be obtainable by claiming a use of a machine learning model that was trained with a specific type of training data.
Conclusions
There is an opportunity for MedTech companies that are incorporating AI tools into their products to seek patent protection for their innovations. Depending on the nature of the invention, a variety of different aspects around may be protectable, ranging from the trained model itself and its use, to the training process and obtaining of training data. Applicants should carefully consider what features of their inventions may be patentable, to ensure that they are effectively protecting their inventions. In many cases, a combination of multiple aspects suggested above can be used to provide strong and commercially valuable protection around the invention.
LET'S TALK
Daniel Brodsky - MedTech & AI Specialist, Patent Attorney, Partner
© 2025 Mewburn Ellis
