Image reconstruction techniques are essential to computer tomography. Algor
ithms such as filtered backprojection (FBP) or algebraic techniques are mos
t frequently used. This paper presents an attempt to apply a feed-forward b
ack-propagation supervised artificial neural network (BPN) to tomographic i
mage reconstruction, specifically to positron emission tomography (PET). Th
e main result is that the network trained with Gaussian test images proved
to be successful at reconstructing images from projection sets derived from
arbitrary objects. Additional results relate to the design of the network
and the full width at half maximum (FWHM) of the Gaussians in the training
sets. First, the optimal number of nodes in the middle layer is about an or
der of magnitude less than the number of input or output nodes. Second, the
number of iterations required to achieve a required training set tolerance
appeared to decrease exponentially with the number of nodes in the middle
layer. Finally, for training sets containing Gaussians of a single width, t
he optimal accuracy of reconstructing the control set is obtained with a FW
HM of three pixels. Intended to explore feasibility, the BPN presented in t
he following does not provide reconstruction accuracy adequate for immediat
e application to PET. However, the trained network does reconstruct general
images independent of the data with which it was trained. Proposed in the
concluding section are several possible refinements that should permit the
development of a network capable of fast reconstruction of three-dimensiona
l images from the discrete, noisy projection data characteristic of PET. (C
) 2001 American Association of Physicists in Medicine.