Signal processing for functional near-infrared neuroimaging

Functional near infrared imaging (fNIRi) is an optical brain imaging modality based on the multi-wavelength absorption spectroscopy using near infrared light with the neuro-physiological underpinning governed by the neurovascular coupling. It is the state-of-the-art in achieving a good balance between the spatial resolution (the two smallest adjacent regions that can be significantly distinguished) and the temporal resolution (the images have to be acquired in real time at a rate comparable to the time evolution of the neural response) while minimally constraining the subject. These factors are crucial in deciding the impact of a brain imaging modality.Signal processing algorithms for optimally processing the recorded data are an essential aspect in the success of any brain imaging technology. Since fNIRi is a relatively young modality (neurovascular coupling was directly confirmed only in 2003), most of the research has concentrated on validation, developing better engineering solutions, improving the SNR and understanding the physics behind the process. This work aims at laying the ground work for a signal theoretic approach for processing fNIRi data. The focus is on developing an understanding of the spectral signature of the signal based on underlying physiology, artifact suppression schemes for cleaning the data, outlier elimination and extraction of features to pave the way for the powerful single trial data analysis schemes.The spectrum of a typical fNIR signal has 4 bands – B waves, M waves, respiration and arterial pulsation (heart beat). The neural response is embedded in the B and M waves bands. The knowledge of the physiological basis of these signals provide a sense of the stationarity of the signal and help in the interpretation of the results; making it vital for the development of any advanced algorithm.fNIR signals are corrupted by three major artifacts – motion, respiration and arterial pulsation and measurement outliers. We present a general adaptive frame work based on the Wiener filter to suppress the motion artifact. The respiration and arterial artifacts are handled by estimating the PSD of the signal of interest which is then used to generate an artifact suppression filter. The Measurement outliers are identified via 2 specifically selected features. The efficacy of all these algorithms is demonstrated on actual data.The robustness of the analysis is directly dependent on the chosen experimental protocol which in turn depends on the information that can be extracted from the recorded signal. We present new features which can potentially be used in single trial analysis – a class of powerful analysis schemes that allow innovative experimental design. As a first step towards the validation of their utility, these features are used in the single trial detection of visual odd ball task. To our best knowledge this is the first attempt at single trial analysis for fNIRi.Together these algorithms constitute a gateway for further development of advanced signal processing schemes for fNIR imaging.M.S., Electrical Engineering -- Drexel University, 200