Three-dimensional detection of fNIRS optodesJenko, Filip (Avtor) Maček Lebar, Alenka (Mentor) Lühmann, Aleksander (Komentor) photogrammetrydetectionfNIRSoptode3D coordinatescolor space HSVcolor stickersalignmentcolor filteringKnowing the 3D coordinates of the optodes during the fNIRS signal recording can be useful for improving the processing of the signals. An atlas of the head can be mapped on the known 3D coordinates which can lead to a better approximation of the anatomical structure of the scanned head, which is needed for the interpretation of the recorded fNIRS signals. The idea of this project was to use a 3D scanner to locate optode – light sources and detectors – end points on the subject’s scalp. Using the landmarks of the head as reference points, the locations of optodes can be given with respect to them. Before scanning, a cap with optodes was placed on the subject’s head. The tops of the optodes and the landmarks of the subject’s head were marked with color stickers, so the 3D coordinates of the optodes and the head landmarks could be determined from the scan. An algorithm to automatically detect the fNIRS optodes was written and verified on human subjects and on a 3D model of a head. Several approaches have dealt with this specific problem already [1], [2], [3], [4], [5], where two of the approaches [1], [4] present a semi-automated method of detection and the optode locations have to be selected manually; in the approach presented in [2], only the landmark reference points are detected automatically and the 3D coordinates of optodes are determined by the predefined ratios of distances between the specific reference point and optode location; another approach [3] presents a pretrained neural network, which predicts the locations of the optodes based from the locations of landmark reference points; the last approach [5], a similar method to the one presented in this thesis is used, where reference points and optode locations are detected based on the color marks on each of them. This thesis and [5] report the longest acquisition times but the best results. In this thesis, the discrepancy to ground truth is given as 1.67 ± 0.62 mm (mean ± standard deviation), and in [5] as 0.5 ± 0.2 mm (median ± median absolute deviation), while the best result of the other papers mentioned was reported by [3] as 3.4 ± 0.9 mm (mean ± standard deviation). This thesis and [5] are the only approaches, in which a 3D printed head model was scanned for calculating discrepancy to ground truth; they are also the only fully automated approaches, where each optode location is detected individually and not based on the predefined distances to landmark locations or determined by a pretrained neural network.20242024-03-14 12:32:31Magistrsko delo/naloga155021sl