The application of AI to medical diagnosis is attracting growing interest from an industrial property perspective. In Europe, these inventions need to overcome two separate hurdles: the exclusion of certain diagnostic methods and the requirements applicable to computer-implemented inventions. This gives rise to particular challenges when they are examined following the criteria of the European Patent Convention.
Over the last few decades, advances in genetic sequencing techniques, genomics, transcriptomics and other omics technologies have enabled the collection of vast amounts of biomedical data, on a scale that would have been unimaginable only a few years ago. However, the generation of data alone does not result in clinically useful knowledge. A biological sample may provide information on thousands of genes, genetic variants, metabolites or biomarkers, but the joint interpretation of all these data poses a challenge that often exceeds human analytical capabilities.
This is where AI models play a particularly important role. Trained with data from previous patients, medical records, medical images, genetic profiles, laboratory results and clinical outcomes, these models can identify patterns and correlations that are not immediately apparent. When the model applies those correlations to data obtained from a specific patient, it can make personalized predictions regarding the patient’s diagnosis, prognosis and even the patient’s response to potential treatments.
The result is an increasingly personalized approach to medicine, where the physicians conclusions are based not only on general clinical parameters, but also on comparisons between the patient’s data and large, previously collected biomedical data. The personalization of the diagnosis and treatment is particularly important in certain fields of medicine such as oncology.
An example of the growing importance of these types of technologies is Roche’s acquisition of PathAI in May 2026. PathAI is a US company that specializes in digital pathology and AI-powered diagnostic solutions. It has developed AISight IMS, a system for managing digital images of tissue based on histological patient samples which, using AI algorithms, helps pathologists evaluate those samples for diagnostic purposes. Roche was already a leading company in diagnostic systems, but believes that the acquisition of these types of AI-based tools will enable it to speed up the discovery of new biomarkers and potential therapeutic targets, thereby increasing its value for biopharmaceutical companies.
However, this trend raises important questions from an industrial property perspective. The inventions used by AI to analyze biomedical data lie at the crossroads of biotechnology and computer-implemented inventions. Such inventions may combine biological elements derived from the human body with computational elements based on mathematical methods, making them particularly complex. This poses a significant challenge for patent agents, since the European Patent Convention (EPC) establishes certain exceptions and exclusions in respect of these elements.
The first hurdle: the exclusion of diagnostic methods
When analyzing an AI-based invention, the focus immediately tends to turn to the rules governing software and mathematical algorithms. However, in the European patent arena, there is a particular aspect that must be taken into account: the exclusion from patentability of diagnostic methods set out in article 53(c) of the EPC. This article was introduced to ensure that patents did not restrict the freedom of physicians and veterinarians to apply the treatment or diagnoses that they considered most appropriate. In particular, it was intended to keep non-commercial and non-industrial medical and veterinary activities outside the scope of patent rights, for ethical, social and public health reasons.
Decision G 1/04 by the Enlarged Board of Appeal of the European Patent Office established that in order to fall within the exclusion from patentability under article 53(c) EPC, the diagnostic method claimed must include the phases constituting the diagnosis: the collection of data, the comparison with values and the attribution of the deviation to a particular clinical picture. In addition, these technical phases must comply with the requirement to be practiced on the human or animal body.
The emergence of AI systems capable of making diagnostic conclusions does not change the logic behind the exclusion from patentability set out in article 53(c) EPC. Decision G 1/04 had already established that the application of this exclusion does not depend on the actual intervention of a physician, but on the nature and structure of the method claimed. Consequently, the relevant criterion is not whether the diagnosis is made by a human brain or an AI system, but whether the patent application’s claim contains the phases that constitute a diagnosis practiced on the human or animal body.
The second hurdle: demonstrating a technical contribution
Once the exclusion in connection with the diagnostic methods has been overcome, the invention faces the rules applicable to computer-implemented inventions. It must avoid falling under the exceptions of articles 52(2)(c) and 52(3) EPC, and in addition must meet the requirements of novelty (art. 54 EPC), inventive step (art. 56 EPC) and industrial application (art. 57 EPC). Broadly speaking, it is insufficient to simply state in the application that a neural network or machine learning model is used, but rather, the patent specification must clearly set out the specific technical contribution made by the claimed solution.
In AI-assisted diagnostic methods, technical character may arise, for example, in the specific manner in which biomedical data are obtained and processed, such as medical images, biomarkers, or genomic and proteomic data obtained from a particular biological sample, as well as from the integration of those data in a technical device or diagnostic support system. The EPO includes some examples in its examination guidelines, such as, the technical character of the use of neural networks in heart monitoring apparatus to identify irregular heartbeats. However, simply obtaining greater diagnostic precision is not enough to prove a technical contribution if the improvement is solely the result of the mathematical model or the data used to train the AI model.
Therefore, claiming that an AI model is capable of making medical predictions with a high degree of accuracy is not enough. Indeed, it is necessary to analyze and include, clearly and concisely in the patent application, the technical problem that the invention resolves, how the data are obtained, the technical features of the model, the nature of the input data used to train it and how it is applied in the medical field.
Conclusion
In its publication Smart Health in 2023, the EPO had already noted that the growth of AI in the healthcare sector had been accompanied by a sizable increase in patent applications, particularly in the field of diagnostic technologies. Bearing in mind that the development of these technologies continues, it seems likely that the number of patent applications in this field will continue to grow in the years ahead. The combination of personalized medicine, large healthcare databases and the rapid development of AI is giving rise to one of the most dynamic areas of innovation in biomedicine today.
In this scenario, the challenge facing innovators is not just in developing models capable of detecting or predicting diseases, but also in identifying which aspect of the solution provides a technical contribution, in order to be able to protect and exploit their inventions. For patent agents the challenge arises in drafting patent applications that protect the invention’s commercial value bearing in mind the specific features of these inventions that lie at the crossroads of biology and computational models.
