Distributed Signal Processing Algorithms for Chronic Neuro-Sensor Networks

Project Information

DISPATCH Neuro-Sense

Grant agreement ID: 802895

DOI

10.3030/802895

Project closed

EC signature date 19 December 2018

Start date 1 January 2019

End date 30 June 2024

Funded under

• EXCELLENT SCIENCE - European Research Council (ERC)

Total cost

€ 1 489 656,00

EU contribution

€ 1 489 656,00

Coordinated by

KATHOLIEKE UNIVERSITEIT LEUVEN
Belgium

Periodic Reporting for period 4 - DISPATCH Neuro-Sense (Distributed Signal Processing Algorithms for Chronic Neuro-Sensor Networks)

Reporting period: 2023-07-01 to 2024-06-30

Summary of the context and overall objectives of the project

The possibility to chronically interface with the brain 24/7 in daily-life activities would revolutionize human-machine interactions and health care. However, such chronic brain computer interfaces (BCIs) will have to satisfy challenging energy and miniaturization constraints. The premise of this project is that these constraints will naturally lead to modular platforms in which multiple networked miniature neuro-sensor modules form a ‘neuro-sensor network’ (NSN).

However, algorithms that extract information from neural signals are designed for traditional neuro-sensor arrays with central access to all data, which makes them unsuited for NSNs where the collected data is distributed over multiple neuro-sensor nodes. Centralizing the data collected by an NSN would require unrealistic energy budgets which are not available in highly miniaturized (battery-powered) neuro-sensors.

The central aim of this project was to eliminate this bottleneck by designing innovative distributed algorithms that enable the nodes of a neural sensor network (NSN) to collaboratively process recorded neural data through in-network data fusion with minimal data exchange. This approach is seen as crucial for leveraging the extensive spatial information captured by NSNs while maintaining feasible energy consumption and bandwidth usage. To achieve this, we developed a novel algorithm design framework that provides a theoretical basis for creating distributed algorithms suited for a wide range of spatial filtering problems commonly encountered in neural signal processing.

For validation purposes, we have mainly focused on establishing these ideas within a new non-invasive NSN concept based on electroencephalography (EEG). We also constructed a hardware demonstrator with wireless miniature EEG sensor nodes for this purpose. By combining multiple of these EEG nodes into an ‘EEG sensor network’ (EEG-Net), we compensate for the lack of spatial information captured by a single node, without sacrificing comfort, flexibility or wearability. We introduced new algorithmic techniques to determine the optimal placement of these sensor nodes, for various BCI and medical applications. The sensors possess local processing capabilities, allowing them to collaboratively process the EEG data using the distributed algorithms mentioned earlier. This represents a significant advancement towards high-performance, chronic EEG monitoring with both a high channel count and low energy consumption.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

1) We have collected a high-density EEG dataset to inform and validate our algorithm design and to gain insights into the properties of signals collected by an EEG-Net. This step is crucial for establishing EEG-Nets as a viable platform for chronic EEG recording, as the modular design influences the signal content due to short electrode distances and galvanic separation between the nodes. To analyze the effects of miniaturization and the absence of a common reference across different sensors, we emulated the sensors of an EEG-Net by re-referencing electrodes to their nearest neighbors. Additionally, to determine the optimal placement of these emulated sensors, we developed automatic electrode selection algorithms tailored for various BCI and medical applications.

2) We have designed innovative distributed algorithms for several key tasks in neural signal analysis, including (1) artifact removal, (2) neural signal enhancement, extraction, and decoding, and (3) spatial feature extraction and classification. These algorithms are crucial for exploiting the spatial information across the modules of a neuro-sensor network (NSN) while maintaining efficient bandwidth usage and energy consumption. The developed algorithms are fully scalable concerning network size and bandwidth requirements. Additionally, these design efforts have given us new insights that have led to the development of a theoretical framework for the design of distributed spatial filtering algorithms. A notable feature of the framework is its ability to automatically generate distributed versions of many well-known centralized spatial filtering algorithms, provided the centralized problem meets certain mild technical criteria. This generation process works for any network topology and ensures both convergence and optimality.

3) We have developed a wireless miniature EEG sensor node that integrates an EEG amplifier, Bluetooth LE radio, processing unit, and battery into a compact package of less than 5 square centimeters. Multiple nodes can be networked and wirelessly interconnected to gather synchronized EEG data from various scalp locations. Additionally, we have designed and integrated dry micro-needle electrodes into the EEG sensor to eliminate the need for gel-based electrodes, making them more suitable for long-term deployment.

4) We have validated these results in the context of various use cases, including brain-computer interfaces and neuro-steered hearing prostheses.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

We have developed a theoretical framework for designing data-driven distributed spatial filtering algorithms that adapt to the dynamically changing correlation structures across all modules in a neuro-sensor network. This framework enables bypassing data centralization, which is typically costly and non-scalable in terms of energy consumption. Algorithms derived from this framework can be proven to achieve performance equivalent to centralized algorithms—essentially as if all modules had access to all data—without the need to share all the data. Instead, each module transmits only a minimal amount of fused data to its neighbors. This approach surpasses conventional distributed parameter estimation algorithms by addressing complex spatial filtering and source separation problems where fused signals (not just parameters) are shared across the network. The framework has been further enhanced with features that relax convergence conditions, enable sensor selection through sparse regularization, improve convergence speed, and reduce computational load, all while maintaining optimality guarantees.

Additionally, our analysis of emulated EEG-Net signals demonstrated that an EEG-Net equipped with galvanically separated miniaturized EEG sensors can achieve the same neural decoding accuracy as standard EEG systems with an equal number of electrodes, given that (a) optimal sensor locations are selected for both setups and (b) the distance between electrodes within each EEG-Net module is at least 3 cm. The latter condition is an empirically determined lower limit for the miniaturization of EEG sensors.

Finally, to the best of our knowledge, our new wireless miniature EEG sensor prototype is the smallest wireless 4-channel EEG sensor to date. It features novel functionality tailored to the EEG-Net concept, which is not available in existing devices. This includes a modular setup where multiple galvanically separated dry EEG sensors work together to collect, process, and share synchronized EEG data within a wireless sensor network.