Home » Posts tagged 'Neural Network'
Tag Archives: Neural Network
Sa-074-Towards an fNIRS foundation model using self-supervision to improve machine learning classification
We propose a novel approach based on self-supervised learning taking advantage of unlabelled segments of functional near-infrared spectroscopy (fNIRS) recordings to pretrain a neural network for classifying n-back levels from brain data. A self-supervised convolutional encoder-decoder was trained on a pretext task consisting in predicting deoxyhaemoglobin (HbR) data from oxyhaemoglobin (HbO), and vice versa. The trained encoder can then be used as a basis for training a subsequent supervised classifier. Ongoing work is investigating this self-supervised approach with more unlabelled data from other datasets to build a robust fNIRS foundation model. Continue reading
Su-019-Voxel-wise modeling of naturalistic auditory stimuli using very-high-density diffuse optical tomography
Language is represented across the cortex in regions known as the semantic system. These regions are extensively mapped with fMRI and provide the foundation for semantic decoding. Here, we establish voxel-wise modeling of narrative stories using semantic features with very high-density diffuse optical tomography (VHD-DOT). Using this model, we predicted voxel responses to a held-out story and validated it against collected data. We found our model to be selective for semantic regions, providing evidence that VHD-DOT can successfully map, and decode narrative stimuli. Continue reading
Su-038-Optimizing spatial specificity and signal quality in fNIRS: An overview of potential challenges and possible options for improving reliability in real-time applications
fNIRS offers great potential for real-time applications such as neurofeedback (NFB) and brain-computer interfaces (BCIs) but faces some challenges. The aim here is to discuss the potential challenges and possible option for improvement, with a focus on increasing spatial specificity and signal quality to enhance the reliability, reproducibility, and repeatability of future fNIRS-based real-time applications. Topics include challenges in probe design, cap placement, and validation using high-resolution techniques such as fMRI. In addition, strategies to improve signal quality, including methods to reduce noise such as motion artifacts and correct extracerebral systemic activity in real-time processing, are discussed. Continue reading
Su-034-Real-time Brain-Computer Interface based on fNIRS
This study proposes a real-time Brain-Computer Interface (BCI) system based on functional Near-Infrared Spectroscopy (fNIRS). We implemented a Rapid Serial Visual Presentation (RSVP) paradigm with stimuli presentation times under one second. Using Lab Streaming Layer, we can record markers of stimuli presentation and participant responses directly onto the fNIRS signal in real-time. We demonstrated the system's ability to distinguish between rare target and non-targets. This preliminary research substantiates the potential for applying fNIRS-based BCI systems into environments requiring rapid stimulus presentation. Continue reading
Su-070-Investigating the consistency of mental task-related fMRI and fNIRS brain-activation networks
We measured the consistency of eight mental task-related brain-activation networks obtained with 3-T fMRI and fNIRS to investigate to which degree these brain-activation maps match, despite the fNIRS’ considerably lower spatial resolution. Results show that brain-activation maps are highly consistent for both HbO and HbR, and largely correspond to statistical maps obtained with fMRI, suggesting that fNIRS is a viable neuroimaging method for mental-task research and its applications. Continue reading