Home » Posts tagged 'AI' (Page 2)
Tag Archives: AI
Fr-021-Investigating the NIRS-derived cytochrome-c-oxidase signal using different system wavelength combinations and processing techniques with a diffusion-based model
The effect of using different numbers and combinations of wavelengths, and varying parameter errors on cytochrome-c-oxidase (CCO) and haemoglobin measurement errors were investigated using a diffusion-based simulation. The number of measurement wavelengths was varied from 5-200, and uncertainties in the differential pathlength factor (DPF) and extinction coefficients of up to 10% were introduced to determine their effect on NIRS-derived concentration changes. Deviations from the true concentration values were found to be below 10% in ideal conditions. However when parameter errors are increased to more realistic values, CCO errors are highly wavelength dependent and orders of magnitude larger. Continue reading
Su-075-The Potential Role of Resting-State Spectral Entropy as a Biomarker in the Progression of Neurodegenerative Diseases
This study applies spectral entropy (SE) to resting-state functional near-infrared spectroscopy (fNIRS) signals of degenerative brain diseases to analyze brain signal complexity in Alzheimer's disease (AD) patients, mild cognitive impairment (MCI) patients, and healthy controls (HC). We show that the complexity variation of brain signals and the synchronization of complexity between brain regions varies with the severity of the degenerative brain disease. We investigate the feasibility of analyzing SE from resting-state signals for future diagnosis of Alzheimer's disease. Continue reading
Su-081-Spatial global component filter vs short channel regression for removing non-cortical component of fNIRS
Non-cortical components can be prominent in fNIRS recordings. These components can be related to blood pressure or respiratory activity and should be removed to measure reliable neural activity or functional connectivity. Here we compare two methods for removing non-cortical components, spatial global component filtering (GCR) and short channel regression (SCR). SCR requires short channel signal acquisition with high signal-to-noise ratio (SNR) and only SNR/(1+SNR) of non-cortical component can be removed. GCR requires large channel coverage and will reduce activity that has large spatial extent. Both methods should be used in appropriate manner. Continue reading
Fr-111-An fNIRS Processing Pipeline Based on Adaptive Thresholding Wavelet Denoising
In this work we present an fNIRS processing pipeline that utilizes BayesShrink based adaptive thresholding in wavelet motion artifact removal. The pipeline is tested on an fNIRS dataset collected from 20 surgical trainees against other methods and achieves significant performance increase (p<0.05) according to both in-group and between group t-test. Continue reading
Su-071-AI-driven large-scale and automated neuroimaging and fNIRS data analysis pipeline using mega-datasets on NeuroJSON.io
We present a proof-of-concept that allows a user to use natural-language instructions to query, retrieve rich neuroimaging datasets hosted on NeuroJSON.io and build data analysis pipelines for large-scale fNIRS data processing. User prompts are converted to machine-actionable commands using a large language model (LLM). We show automated anatomical scan segmentation, 3-D mesh generation and mesh-based Monte Carlo simulations, but this framework can readily be expanded to other types of analyses. Continue reading
Su-069-Localizing fNIRS hemodynamic oscillations on the cortical surface using wavelet Maximum Entropy on the Mean
Reconstructing fNIRS hemodynamic fluctuations on the cortical surface is an ill-posed problem that requires validation. The wavelet-based maximum Entropy on the Mean (wMEM) method has recently been validated for the localization of EEG/MEG resting-state activity. In this work, we adapted wMEM for the localization of fNIRS resting state activity, validating it using controlled simulations and localizing hemodynamic fluctuations during sleep. Continue reading
Vi-029-Shedding light on Fibromyalgia: Developing a Machine-Learning-Based Diagnostic Tool Using fNIRS
This study develops a machine learning-based tool for diagnosing fibromyalgia (FM) using functional near-infrared spectroscopy (fNIRS). We recorded cortical activity from 36 FM patients and 33 age-matched controls, focusing on frontal and sensory-motor areas. Data was processed to remove artifacts and analyzed for connectivity and task-based features during rest and during stress. A convolutional neural network, trained on 85% of our dataset, achieved an 89% accuracy rate in diagnosing FM, identifying all FM patients accurately while misdiagnosing one control. This biomarker predominantly reflects frontal cortex activity, offering clinical utility and insights into the role of stress mechanisms in FM. Continue reading
Fr-110-Impact of neurostimulation on performance enhancement in robotic surgery
Robotic surgery has enhanced operation flexibility and patients’ recovery while extending surgeons’ training time over traditional methods at the same time. Although tDCS neurostimulation improves immediate motor behaviour, its impact and long-term retention on cortical dynamics are not understood. fNIRS data was longitudinally collected from a cohort of surgeons during a robotic surgery knot-tying tasks receiving sham vs tDCS neurostimulation. This study applies a GLM to examine differences at the fNIRS data during the pre-, post- and retention sessions. This analysis aims to reveal the mechanism underlying tDCS’s influence on robotics surgical skills and motor learning. Preliminary results suggest tDCS leads to a significant activation in the right hemisphere. Continue reading
Su-092-Towards automated bad channel detection in functional near-infrared spectroscopy
With more channels per device, larger cohorts, and the need for standardized processing procedures, thresholding-based detectors are becoming widely employed in fNIRS to identify unreliable signals. However, despite their potential, un- and semi-supervised detection have yet to be utilized for bad channel detection, and a comprehensive assessment is lacking. We developed three novel detectors and compared them with established methods, un- and semi-supervised techniques. Machine learning approaches that utilize partially rated data, particularly the here developed NiReject, achieved superior performance. This may pave the way for improved, automated quality control in fNIRS and artificial intelligence-based developments for such control. Continue reading
Sa-056-Broadband NIRS reconstruction with colouration maps
Colourations Maps offer a look-up table solution to inverse problems. Their reconstruction error is limited by the entries in the table in turn limited by computational demands. Crust & Crumb, a pseudo-random gradient-weighted sampling design is put forward aiming to escape the combinatorial explosion of traditional sampling designs whilst keeping the reconstruction error at bay. A synthetic exponential model alike the Beer-Lambert law is constructed to establish the validity of the new sampling design, and a finite element diffusion model exemplifies its feasibility against spectroscopic data. In experiments, Crust and Crumb exhibited superior performance over alternatives. Continue reading