1 Multimodal Functional Imaging Laboratory, Biomedical Engineering Department, McGill University, 3801 Rue University 751, Montreal, QC, H3A2B4, Canada. alexis.machado@mail.mcgill.ca.
2 Multimodal Functional Imaging Laboratory, Biomedical Engineering Department, McGill University, 3801 Rue University 751, Montreal, QC, H3A2B4, Canada.
3 Department of Physics and PERFORM center, Concordia University, Montreal, Canada.
4 Department of Neurology and Neurosurgery, Montreal Neurological Institute, McGill University, Montreal, Canada.
5 École de technologie supérieure de l'Université du Québec, Montreal, Canada.
6 Centre de Recherches en Mathématiques, Montreal, QC, Canada.
7 Centre d'Etudes Avancées en Médecine Du Sommeil, Centre de Recherche de l'Hopital Sacré-Coeur De Mo

In functional near infrared spectroscopy (fNIRS), deconvolution analysis of oxy and deoxy-hemoglobin concentration changes allows estimating specific hemodynamic response functions (HRF) elicited by neuronal activity, taking advantage of the fNIRS excellent temporal resolution. Diffuse optical tomography (DOT) is also becoming the new standard reconstruction procedure as it is more accurate than the modified Beer Lambert law approach at the sensor level. The objective of this study was to assess the relevance of HRF deconvolution after DOT constrained along the cortical surface. We used local personalized fNIRS montages which consists in optimizing the position of fNIRS optodes to ensure maximal sensitivity to subject specific target brain regions. We carefully evaluated the accuracy of deconvolution when applied after DOT, using realistic simulations involving several HRF models at different signal to noise ratio (SNR) levels and on real data related to motor and visual tasks in healthy subjects and from spontaneous pathological activity in one patient with epilepsy. We demonstrated that DOT followed by deconvolution was able to accurately recover a large variability of HRFs over a large range of SNRs. We found good performances of deconvolution analysis for SNR levels usually encountered in our applications and we were able to reconstruct accurately the temporal dynamics of HRFs in real conditions.