Graduation Semester and Year




Document Type


Degree Name

Master of Science in Biomedical Engineering



First Advisor

Georgios Alexandrakis


Functional near infrared spectroscopy (fNIRS) is a technique that enables monitoring brain function, by detecting changes in near-infrared light absorbance resulting from changes in brain hemodynamics during periods of rest and activation. Although the technique has some advantages over other conventional functional imaging methods with its inherently high temporal sampling and detection sensitivity, the sparse spatial sampling of current fNIRS systems limits spatial resolution and requires help from Magnetic Resonance Imaging (MRI) to attain spatial co-registration with anatomical structures. In this work we have performed simulations to test the feasibility of using a high density spatial sampling system as a means to detect anatomical structures on the brain surface using near-infrared light signals only while also significantly improving the efficiency of signal detection. We chose to perform our simulation using a Monte Carlo eXtreme code, which is an accelerated GPU-based Monte Carlo simulation technique that resulted in computational time gains of 300X over a traditional code on a non-GPU set-up. This acceleration is achieved due to high thread parallelization and improved memory latency that speeds up the algorithm to trace multiple simulated photons in parallel. With the vision to create a simulation set-up that mimics reality as much as possible, we sourced an anatomical brain model from MRI and included hemodynamic fluctuations in different tissue compartments as explained below. The 3-D image volume was spatially processed in Statistical Parametric Modeling software and subsequently in MATLAB to transform it into a simulated tissue model. The simulation set-up was designed to place a dense 1mm grid of detector fibers on the scalp. Tissue optical properties were defined at the common fNIRS wavelength of 830 nm. An estimate for an adequate simulated photon number to make the stochastic noise from Monte Carlo smaller than the amplitude of hemodynamic fluctuations was made by running independent trials on the tissue model and analyzing the standard error between trials. A resting state brain model was considered to be appropriate for testing the feasibility of detecting cortical sulci by high density fNIRS since background hemodynamics are known to be present during all times. In order to create the hemodynamic background, information of common sources of physiological hemodynamics, namely Mayer waves (~0.1Hz) and respiratory waves (0.2 - 0.4 Hz) was sourced from the fMRI BOLD data coregistered with the anatomical MRI image volume used previously. The fMRI dataset had lower temporal sampling (0.5 Hz) and hence only Mayer waves and not respiratory waves could be sourced from the fMRI data. Respiratory waves from an fNIRS baseline data set sampled at 10.35 Hz were introduced into the brain tissue model at the scalp and gray matter in a depth-independent manner. The power spectra of hemodynamic fluctuations from Mayer waves and respiration were combined into a consistent single power spectrum and were added to the brain tissue voxels. The resulting hemodynamic fluctuation data were used to modify the light absorbance simulated by Monte Carlo and create resting state time-series reflectance data for the simulated high density fNIRS system. Four simulations for sources at the corners of a 27 mm sized source paired with detectors along a circle of 24 mm radial distance were chosen as the best geometry for reconstructing 2-D cortical hemodynamic fluctuation maps. We analyzed the resulting images by a signal cross-correlation method and were able to identify cortical sulci from gyri within the center of the imaging field of view. Interestingly this approach could detect sulci that were even 2mm deep form the cortical surface. These preliminary results show that it may be worthy building a very high density fNIRS for mapping anatomical features along with activation maps.Furthermore, to quantify the benefits of dense spatial sampling on signal collection, a simulated activation region was embedded into the central sulcus of the sensorimotor region in the cortex. Appropriate hemodynamic response functions for this activation region were designed for finger tapping at 1 Hz for 16 sec, 8 sec and 4 sec. Detector fibers in the proposed system were grouped to determine an effective detector diameter size of 13 mm as the most appropriate for maximizing the activation signal-to-noise ratio of activation. Compared with sparser spatial sampling from a conventional fNIRS system, the high density system offered gains of 125% - 400% in signal-to-noise ratio depending on detector placement with respect to the activation location. Also, the dense spatial sampling system showed prospects of reducing the total duration of an activation protocol by half. Finally, photon budget calculations demonstrated the feasibility of collecting adequate signal from a single detector fiber while staying within light power exposure safety limits, which would have to be taken into account in a real life system. The simulation feasibility studies performed here show that a high density sampling systems holds potential for revolutionizing the fNIRS field.


Biomedical Engineering and Bioengineering | Engineering


Degree granted by The University of Texas at Arlington