Brainwave to breakthrough: EEG data as Intellectual Property

Introduction

In recent years, a growing field of research has focused on the intersection of brainwave monitoring and machine learning. Electroencephalogram (EEG) signals—recordings of brain activity—have long been used in medical settings to assess conditions such as Parkinson’s disease, dementia, epilepsy, stroke, and sleep disorders. But their utility extends far beyond healthcare. EEG technology is now being harnessed in wellness industries, where it offers insight into consciousness and well-being, as well as in more cutting-edge applications such as neuromarketing, augmented reality, toys, gaming, and even lie detection.

With the ability to capture and analyse EEG signals using advanced machine learning techniques, including neural networks, deep learning models and regression models, researchers are discovering new ways interpret brain activity and predict states. This wealth of data presents a unique opportunity to create valuable intellectual property (IP) in the field. In this article, we explore how the growing trend of EEG data analysis and modelling could lead to significant innovations and protectable IP.

Patent protection

In Europe, a wide range of innovations related to the capture and processing of EEG signals can be considered for patent protection. Key areas for potential patents include:

• (i) Mechanical or electronic aspects of data capture: This includes innovations related to EEG headsets, electrodes, amplifiers, digital signal processing and other devices and methods involved in gathering EEG signals or data.
• (ii) Machine learning models and pattern identification: Patents can cover data processing techniques and purposeful method using models developed through training on EEG data sets, capturing the advanced algorithms and methods used to detect a disorder.
• (iii) Predictive EEG parameters: This involves patents for identifying specific EEG parameters or combinations thereof that are highly predictive of certain diseases, disorders or statuses, as determined by the machine learning model.

Each of these categories represents a significant area of potential IP protection in the field of EEG technology. We will briefly explore each of these categories below.

Category (i)

The first category of EEG-related innovation typically falls within the realms of mechanical engineering, material science, electrical engineering or digital signal processing. This includes inventions that capture, process and store EEG data, such as for example wearable EEG devices. Innovations in this space might involve the design of electrode supports, such as caps, headbands, or even smart clothing, as well as advancements in the electrodes themselves—whether through new conductive materials or gels. Other developments could focus on the overall portability and wearability of the system, the arrangement of electrodes, improvements in the electronics such as more sensitive or energy efficient amplifiers, and improvements in dedicated software such as faster or more efficient digital signal processing techniques.

These types of inventions belong to a well-established and relatively predictable patent domain, with harmonisation across most global patent systems. Unlike some medical methods, which face exceptions from patentability in Europe and eligibility issues in the U.S., innovations in this field are generally considered strong and reliable patent assets.

Category (ii)

A growing focus for innovators in EEG-related technology is on extracting useful information from EEG data. Such innovators often turn to another layer of protection: pattern identification techniques or machine learning models trained on EEG data. This approach allows for an application of the machine learning model to be patented by specifying its unique mathematical terms and weights—factors that result from meticulous training on curated data. Developing these models involves extensive trial and error, including creating pre-processing protocols, choosing the right model, and fine-tuning through training and validation. Among these, the model's type and weights are often the most valuable and sensitive component. It also allows for any other data processing techniques to be patented by specifying their functionality, typically in the form of method steps. When a data processing technique is able to identify a new pattern, which identifies a disease, or if it is able to generate key biomedical parameters in a more reliable and efficient manner than prior art techniques, chances of obtaining a patent might be high.

However, patenting a method making use of the model type and weights  or the data processing method requires public disclosure, which companies are sometimes reluctant to do. To avoid this risk, an alternative option is to protect this information as a trade secret instead. In the European Union, trade secrets are legally recognised and protected against unlawful acquisition. More on this protection strategy is explored in the Trade Secret section below.

Category (iii)

For applicants who find direct protection of a machine learning model and its weights or a data processing method and its algorithmic details off-limits, there is still potential to secure valuable IP by focusing on other aspects of the trained model or the data processing method. Specifically, patterns identified during the training process or by the data processing method, using large EEG datasets and a defined ground truth—such as a medical condition—can be claimed without disclosing the model or the data processing method itself. In such case, certain input parameters derived from EEG signals become critical. For instance, when using EEG data to predict Parkinson’s disease, parameters such as measurement location, frequency bands, or signal-processed features might be extracted from the EEG stream. These parameters – often in the 100’s – combined with the ground truth (e.g., Parkinson’s disease: Yes/No), train the model.

Through this training process, it becomes clear which parameters are the most predictive and which hold little significance. If these predictive parameters are novel and unrecognised in prior art, there is an opportunity for a patent. Applicants can claim the most predictive parameters, sometimes referred to as "digital health markers," within context of a purposeful method, often specifying how changes in their values influence the likelihood of a condition. This strategy works best when the predictive parameters are rankable, because it allows a claim reciting only a few of the most significant digital health markers as opposed to the full training set.

Even when input data cannot be easily broken down into specific parameters, there may still be room for patent protection if the data belongs to a broad category and no prior art links that category to the ground truth. For example, imagine a data processing method, or machine learning model trained on retinal images that reveals a strong correlation to EEG signals, essentially serving as an EEG proxy. This novel association could support a claim for predicting EEG signals or related disorders based on retinal images. Such an approach opens the door to innovation, leveraging broader data types to establish connections previously unexplored in the field.

EPC exceptions for diagnostic methods

In Europe, claims directed at diagnostic methods are restricted by Article 53(c) of the European Patent Convention (EPC), which prohibits patents for “...diagnostic methods practised on the human or animal body.” As a result, a method claim aimed at predicting a disorder—such as Parkinson’s disease—based on EEG signal parameters would likely face objection under this provision.

However, there are ways to overcome this constraint. While diagnostic methods themselves are excepted from patentability, product claims are not. A claim can be reformulated to focus on a medical device, computer program product, storage medium, or another type of product. 

Other potential ways to overcome the objection might include:

• Indicate that the EEG data was collected in the past, distancing the claim from a real-time on-the-body method.

• Reformulate the claim as a method for obtaining information, as outlined in the EPO Guidelines (G-II 4.2.1.3), rather than directly diagnosing a condition. 

• Avoid any steps that involve making a diagnosis (case law G1/04, step iv), and instead frame the method as one for measuring the severity of symptoms or other related metrics of an already-diagnosed condition. 

EPC exceptions for surgical methods

Claims directed at surgical methods are also ruled out by the same Article 53(c) EPC, which prohibits patents which involve a bodily invasive step representing a substantial physical intervention on the body which requires professional medical expertise to be carried out and which entails a substantial health risk even when carried out with the required professional care and expertise (G1/07). This can invoke a further objection when the predicting method specifies a particular variety of EEG electrode that is an implantable EEG electrode or signals obtain therefrom. Again, a robust solution would be to reformulate the method claim as a product claim.

Trade secret

In the context of protecting trained machine learning models and their weights, trade secrets offer an alternative to patents. According to EU Directive 2016/943, a trade secret is information owned by a company that is secret, that has commercial value owing to its secrecy and wherein reasonable steps are taken by the company owning the trade secret to keep it secret. Reasonable steps to keep the information secret are a basis for establishing a trade-secret policy that is key to protecting such sensitive data. Some best practices include:

• Limiting access to the trade secret strictly to those who need to know.
• Implementing robust IT security measures to prevent unauthorized access, including restrictions on personal devices and file sharing, as well as tracking access to sensitive information.
• Conducting thorough onboarding and exit interviews to ensure employees are aware of their obligations concerning confidential data.
• Providing regular training to reinforce understanding of what constitutes confidential information and treating trade secrets as part of compliance.
• Utilizing non-disclosure agreements (NDAs) with third parties and incorporating confidentiality clauses into contracts to further protect trade secrets.
• Enforcing proper contract management, ensuring non-compete and non-solicitation clauses, and monitoring contract expiration and renewal.
• Using non-public iDEPOT depositions to document the existence and content of a trade secret at a specific point in time.

A trade secret is protected by law against theft according to the directive. However, it does not protect against a third party independently developing the same machine learning model. This is where patents offer advantages, as they can prevent others from using the same innovation, even if developed independently. 

Like patents, trade secrets do not automatically grant freedom-to-operate, meaning someone with a protected or secret model could still face legal challenges from third party patent holders. However, in some jurisdictions, prior user rights offer some safeguards, allowing established users of a trade secret to continue using it even if a patent is later issued to someone else for similar technology, helping shield them from contentious actions by patent holders.

Prior user rights

The principle of prior user rights allows a party that has been using or preparing to use a technology (e.g. maintained as a trade secret) to continue using that technology even if a third party later files a patent for it. This protection is jurisdiction-specific and requires that the prior use occurred within the patent-protected territory before the filing of the patent application, and that any subsequent use is the same as the prior use.

For example, if a company has been using a trade secret in Belgium and a competitor later files a patent application in the U.S. for that same technology, the company’s prior use in Belgium would not protect its later use of the technology in the U.S. because the prior use did not occur in the U.S. (it was post-use). However, the company's activities in Belgium could continue under the principle of prior user rights, as long as the prior use occurred before the patent was filed in Belgium. In essence, prior user rights can act as a shield, but only within the specific territory where the technology was in use prior to the patent filing.

Stakeholders need to be aware that the prior user right is limited to the model or software in use before the filing of the third party patent. This will seldom be identical to the model or software in use a few years later when the third party patent gets granted. 

Design rights

When patent protection under Category (i) is unavailable due to prior art, or in supplementation of patent protection, a further viable strategy is to seek a design right, which safeguards the visual appearance of a product that isn’t purely utilitarian. A design right protects the product’s aesthetic features, such as its lines, contours, colours, shapes, textures, and materials, or the ornamentation of the product. This can apply to any object manufactured through industrial or traditional methods, including components meant to be assembled into a larger product, packaging, graphic symbols, and even typography.

A design right may cover either the three-dimensional aspects of the product itself or the two-dimensional ornamental elements. However, design protection is limited by the "must-fit" exception, which excludes designs whose appearance is dictated entirely by their technical function or the necessity to interconnect with other products to serve a technical purpose. This option provides an additional layer of IP protection, focusing on the product’s aesthetic value rather than its function.

Trademarks

Another avenue for IP protection could emerge through the creation of novel labels for EEG parameters or outputs, aimed at establishing market benchmarks. For instance, if EEG data were found to quantify an athlete's state of high focus, a sign could be coined to describe this state. The chosen term, if widely adopted in research, commercial products and usage, could gain traction as a market reference, which will make it attractive for competitors to adopt the use of such a sign or similar signs. This may then disrupt your business due to the likelihood of confusion caused or the unfair advantage taken from your marketing efforts.

The most easy way to legally defend against such scenarios is to protect this sign, including its verbal form and potentially its visual representation, via a trademark. Traditionally, trademarks encompass names, phrases, logos, and symbols, but can also extend to more unconventional elements like shapes, colour, sound, patterns or positions. Such trademarks could play a pivotal role in raising market awareness and maintaining brand loyalty.

Be aware however that trademark protection is reserved for signs that are distinctive and therefore do not describe the goods and services for which they are being registered. It is also recommended to conduct availability searches before adopting or protecting a sign in order to avoid later conflicts with older, similar trademarks.

Conclusions

EEG signals are a dynamic and commercially promising area of innovation, fuelled by the growing availability of EEG data and the accessibility of machine learning technology. Patent protection can cover various aspects, including the tangible devices (medical devices, gaming headsets), processing methods, machine learning models, and digital markers (e.g. health markers, motion markers). Beyond patents, companies can also leverage EU trade secret directive to protect their trained machine learning models from theft, explore design rights for the visual appearance of their products, or consider trademarks to christen novel EEG parameters or outputs. 

For expert guidance on the most effective strategies to safeguard your innovations, please contact us at info@dcp-ip.com. We are here to guide you towards the best ways to protect your valuable IP.