Using an image database to constrain the acquisition and reconstruction of MR images of the human head

A training set of MR images of normal and abnormal heads was used to derive a complete set of orthonormal basis functions which converged to head-like images more rapidly than Fourier basis functions. The new image representation was used to reconstruct MR images of other heads from a relatively small number of phase-encoded signal measurements. The training images also determined exactly which phase-encoded signals should be measured to minimize image reconstruction error. These signals were nonuniformly scattered throughout k-space. Experiments showed that head images reconstructed with the new method had less serious truncation artifacts than conventional Fourier images reconstructed from the same number of signals. The resulting images were characterized by spatially variable spatial resolution and were particularly well-resolved in regions where the training images had structural detail.