PhD Scientific Days 2023

Neurosciences - Posters H

Improving ECoG recordings with missing electrodes for classification through deep learning

Text of the abstract

Electrocorticography (ECoG) is an invasive technique used for recording electrical activity from the brain. It offers high-resolution recordings of brain activity, making it essential in certain clinical cases. However, the placement of electrodes requires an invasive process, and the electrodes are often placed based on the specific needs of each patient, which can result in suboptimal placements or missing electrodes. In this study, we propose a GAN-based approach for inferring missing electrodes in ECoG recordings to improve the spatial resolution of the recordings.

The proposed approach consists of two steps: inferring missing electrode signals using a GAN, and classifying the mixture of inferred and real electrode signals. The GAN generator network is pre-trained on augmented data, where ground-truth signals are available for the sensorimotor cortex (SMC), and then applied to recordings with missing or sparse signals from the SMC. The generator network takes the existing ECoG electrode recordings as input and generates a complete electrode grid, while the discriminator network evaluates the generated grid to determine whether it is real or fake.

The inferred electrodes are then classified using a separate deep learning classifier network to determine whether the subject moves their hand or is at rest. Our proposed approach has the potential to improve the spatial resolution of ECoG recordings, providing a more complete picture of the brain. Additionally, this approach could enhance the training of ECoG-based brain-computer interfaces. Our ongoing work aims to evaluate the effectiveness of the proposed approach.

Overall, our study highlights the importance of considering missing electrodes in the analysis of ECoG recordings and demonstrates the potential of deep learning approaches for improving ECoG recordings with missing electrodes. 

Project no. FK132823 has been implemented with the support provided by the Ministry of Innovation and Technology of Hungary from the National Research, Development and Innovation Fund, financed under the FK_19 funding scheme. This research was also funded by the Hungarian Brain Research Program (2017_1.2.1-NKP-2017-00002) and the TUDFO/51757-1/2019-ITM grant by the Hungarian National Research, Development and Innovation Office. Project no. RRF-2.3.1-21-2022-00015 has been implemented with the support provided by the European Union. JR is thankful to Semmelweis University for the EFOP-3.6.3-VEKOP-16-2017-00009 grant.