Training data from: Reconstructing cerebral hemodynamics from sparse data using Neural Operator Transformers

Dataset used for training GNOT and VANO models as described in: Kaczmarek W, Magdziarz Ibrahim-El-Nur J, Łoś M, Roleder T. Reconstructing cerebral hemodynamics from sparse data using Neural Operator Transformers. Comput Biol Med. 2025 Sep;195:110492. doi: 10.1016/j.compbiomed.2025.110492. Epub 2025 Jun 18. PMID: 40554056. Dataset consists of solutions to 1D ROM blood flow equations solved on arterial network of Circle of Willis (CoW), with accompanying simulation parameters in "theta" vector ( radii and lengths of all arteries in particular CoW instance, heart rate, total resistance and compliance, one-hot encoding indicading artery). Each entry in dataset represents solution on single artery. All simulations were run for 10 cardiac cycles to ensure convergence to a periodic state, verified by checking that the relative difference between waveforms in the last two cycles was below 1%. Additional details and ranges for parameter sampling can be found in the manuscript. Dataset is saved in .npy file with following format: Dataset is saved in .npy file with following format: Dataset = [ [X1, Y1, Theta1, Inputs_funcs1], [X2, Y2, Theta2, Inputs_funcs2], ... ] X: (N x N_in) numpy array, representing input mesh points N : number of points N_in: input spatial dimension Y: (N x N_out) numpy array, representing physical fields defined on these mesh points N_out: output dimension Theta: (N_theta,) numpy array, global parameters for this sample N_theta: dimension of global parameters Input_funcs: tuple (inputs_1, ..., inputs_n),every inputs_i is a numary array of shape (N_i, f_i): N_i: number of points used to discretize this input function f_i: dimension of this input function plus the dimension of geometry, actually it is the concat of (x_i, f(x_i)).