Image reconstruction by a Hopfield neural network

Abstract

The reconstruction of cross-sectional images from projections involves the solution of a large system of simultaneous equations in which the unknowns are attentuation coefficients associated with the cells consti tuting the image. As an alternative to iterative methodssuch as the algebraic reconstruction technique (ART), a modified linear form of the Hopfield neural network with a summation layer to significantly decrease the number of interconnections is proposed. Higher speed and improved SNR, compared to ART, have been obtained on the Shepp and Logan 'head phantom' divided into 100 × 100 cells. © 1993.