A blood vessel extraction apparatus is characterized by a convolutional neural network (600) including a convolution unit (600a) which has a1: a first path (A1) to which, data (51) for a first target region (R1) including some target voxels (681a) in the medical volume data is input at a first resolution, and a2: a second path (A2) to which, data (52) for a second target region (R2) including some target voxels (681a) in the medical volume data is input at a second resolution, and b: an output unit (600b) having a neural net structure, which outputs numerical values related to visualization of the target voxels with an output result of the first path and the second path as input data.
