Apparatuses and a method for artifact reduction in medical images using a neural network

An apparatus, comprising:

• processing circuitry configured to train a neural network by
• obtaining a pair of reconstructed computed tomography (CT) images including an artifact-exhibiting image and an artifact-minimized image, the artifact-exhibiting image having a greater degree of artifacts than the artifact-minimized image, including
• obtaining the artifact-minimized image by reconstructing the artifact-minimized image from first projection data, the first projection data being radiation data obtained using an X-ray beam that is less than a predefined angle threshold, and
• simulating the artifact-exhibiting image from the artifact-minimized image, including forward projecting, using a CT scan configuration with the X-ray beam that is greater than the predefined angle threshold, the artifact-minimized image to generate second projection data from which the artifact-exhibiting image is generated;
• forming a training dataset including the pair of reconstructed CT images;
• applying the artifact-exhibiting image to the neural network to generate an output image having reduced artifacts with respect to the artifact-exhibiting image, the neural network including neuronal nodes connected by connections having weighting coefficients;
• calculating a cost function representing a difference or disagreement between the output image and the artifact-minimized data;
• updating the weighting coefficients in the neural network to optimize the cost function;
• and
• stopping, upon satisfying a predefined stopping criteria, the updating of the weighting coefficients, so that the neural network is then a trained neural.