Home » Posts tagged 'Signal Processing' (Page 4)
Tag Archives: Signal Processing
Fr-035-Subjectivity and Brain Activity-Based Validation to Driving Feedback
In this study, we aim to realize a real-time feedback system that encourages drivers to continuously become aware of safe driving. There are various feedback methods, and so far, five types of subjective impressions have been evaluated using images, warning sounds, and announcements. However, as subjective evaluation alone may lead to responses that intentionally or unintentionally influence the outcome of the experiment, we focused on the brain, which is processing the information, and evaluated feedback by measuring brain activity with NIRS. The results of the experiment indicated that the brain may be activated to process information for those with high subjective evaluation content. Continue reading
Fr-049-Cedalion: A Python-based framework for data-driven analysis of multimodal fNIRS and DOT
We would like to announce the release of a new Python-based software framework, Cedalion, designed for data-driven fNIRS/DOT analysis, with the ultimate goal of merging well-established analysis methods for neuroimaging modalites (e.g. fNIRS, EEG) and physiological and behavioral measurements, with advanced multimodal machine learning techniques. Continue reading
Fr-047-A Multi-Atlas Machine Learning Approach for Automated Segmentation of Widefield Optical Imaging in Mice
In neuroimaging studies, precise brain and non-brain pixel labeling is essential for robust processing and statistical testing. Traditionally, such segmentation of widefield optical imaging data is performed manually, leading to high inter-rater variability. Current state-of-the-art machine learning approaches use a single baseline false-color or fluorescence image and are not generalizable to multi-institutional data. Our approach maintains data from all illumination wavelengths, using convolutional networks for segmentation. Multi-atlas techniques exhibited statistically significant improvements over single-atlas segmentation, showing promising results for efficiency and reduced variability. Future endeavors will refine models and expand datasets for broader applicability. Continue reading
Fr-079-Multivariate Hierarchical Noise Models of fNIRS Data
In this work, we present justification for why multi-level noise modeling is needed in fNIRS analysis and how miss-matched task-dependent noise (single-trial variability, measurement errors, motion-artifacts, etc.) can result in crosstalk and under/over-estimation of statistical effects sizes between estimates of task results. We propose a new multivariate, hierarchical GLM analysis. Continue reading
Fr-078-Brain Predictive Text Decoding using fNIRS
In this work, we introduce a novel brain decoding scheme utilizing a brain-to-text model developed by coordinate-based meta-analysis (CBMA) of existing functional MRI literature. In this approach, which has been previously demonstrated for the interpretation of fMRI results, a multivariate model linking reported brain activity coordinates converted to international 10-20 space is statistically linked to words used in the publications associated with those coordinates. The resulting model allows us to generate statistical word associations (word-clouds) from an image of fNIRS brain activity projected to 10-20 space. Continue reading
Fr-099-Enhancing Within-Subject Consistency of fNIRS Metrics: Regression of Physiological Contamination Using Short-Channels and Auxiliary Signals
Short-separation (SS) channels are commonly used to remove physiological interference from fNIRS data, but their use in single-trial tasks and less common metrics like functional coherence (FC) is limited. Our study involved healthy controls attending three visits a week apart, where physiological vital signs and fNIRS responses during rest and non-rest tasks were quantified. We developed a SS-channel regression algorithm to include vital sign effects in the regression model and assessed within-subject consistency before and after regression on post-processed data (e.g., FC). This approach aims to enhance fNIRS data consistency, and potentially improve fNIRS-based classification models and research outcomes. Continue reading
Vi-006-Channel Selection using Ant Colony Optimization for fNIRS Based BCI
The availability of fast and accurate fNIRS signals is crucial for the rehabilitation of disabled individuals. Signal processing and classification can be expedited and enhanced in accuracy by selecting channels with higher activation levels. In this study, we apply a meta-heuristic algorithm, Ant Colony Optimization (ACO), to fNIRS signals for channel selection. We compare the results with state-of-the-art methods such as the t-value and z-score methods. Our findings demonstrate that utilizing ACO for channel selection can significantly improve classifier accuracy. Continue reading
Vi-031-Motion Artifact Correction in preschoolers’ data: Comparison of five pipelines
Abstract: This study investigates motion artifact correction methods in preschoolers’ fNIRS data. Using QT-NIRS in the pre- and post-correction, we compared five different motion artifact correction methods using real data in preschoolers at the age of 2.5-4.5 years. We found significant improvements using Wavelet and Wavelet+TDDR corrections over the other methods. This study is the first to compare various pipelines for preschoolers’ data and to use QT-NIRS as an objective comparison of motion artifact correction methods. Continue reading
Fr-109-Individualized head models from registered optodes or photogrammetry for improved DOT image reconstruction
We propose a data-driven algorithm to approximate individual head anatomies in case of unavailable MRI/CT scans to improve image reconstruction accuracy over widely used standard head models such as Colin27. Based on a low-dimensional representation of a large head model database we derive individual head parameters solely from additional knowledge of the subject’s scalp as gained from e.g. photogrammetry scans or exact optode positions. Preliminary photon simulation results show significantly less deviation from the subject’s ground truth head model compared to Colin27 and raises hopes for improved source reconstructions with little effort in many future fNIRS studies. Continue reading