Abstract: Tomographic images, acquired by iterative reconstruction of lower quality images, are enhanced by a trained neural network. Next, the enhanced tomographic images are input to the next step of the iterative reconstruction. For this purpose, one or several neural networks are trained with a first set of tomographic images and a second set of tomographic images at lower quality. The second set of tomographic images at lower quality are acquired by applying an iterative reconstruction algorithm to lower quality projection images. The iterative reconstruction can use a normal quality tomographic image as input.