Device for inferring material density image, CT system, storage medium, and method of creating trained neural network

A CT system for collecting single energy CT data, comprising:

• an X-ray tube in which a prescribed tube voltage is applied;
• and
• one or more processors, wherein
• the one or more processors performs an operation, the operation including:
• generating a CT image based on single energy CT data collected from a subject body;
• inputting the CT image into a first trained neural network,
• the first trained neural network being created by a first neural network performing learning using a plurality of virtual monochromatic X-ray images in a training phase;
• causing the first trained neural network to infer a virtual monochromatic X-ray image based on the CT image;
• generating a first material density image expressing the density of a first reference material and a second material density image expressing the density of a second reference material based on the CT image and the virtual monochromatic X-ray image inferred by the first trained neural network;
• inputting the first material density image and second material density image into a second trained neural network,
• the second trained neural network being created by a second neural network performing learning using a plurality of material density images in a training phase;
• and
• causing the second trained neural network to infer a third material density image expressing the density of the first reference material and a fourth material density image expressing the density of the second reference material based on the first material density image and second material density image.

A method of creating a trained neural network, comprising the steps of:

• creating a first trained neural network,
• the first trained neural network being created by a first neural network performing learning using a first training data set,
• the first training data set containing a first image set containing a first plurality of virtual monochromatic X-ray images and a second image set containing a second plurality of virtual monochromatic X-ray images,
• each virtual monochromatic X-ray image of the first plurality of virtual monochromatic X-ray images being a virtual monochromatic X-ray image of a first energy level corresponding to a tube voltage of a CT system that collects single energy CT data, and each virtual monochromatic X-ray image of the second plurality of virtual monochromatic X-ray images being a virtual monochromatic X-ray image of a second energy level, and
• the first neural network performing learning using the first training data set, such that each virtual monochromatic X-ray image of the first plurality of virtual monochromatic X-ray images is used as input for the first neural network and each virtual monochromatic X-ray image of the second plurality of virtual monochromatic X-ray images is output from the first neural network;
• and
• creating a second trained neural network,
• the second trained neural network being created by a second neural network performing learning using a second training data set,
• the second training data set containing a third image set containing a first plurality of material density images expressing the density of a first reference material, a fourth image set containing a second plurality of material density images expressing the density of a second reference material, a fifth image set containing a third plurality of material density images expressing the density of the first reference material, and a sixth image set containing a fourth plurality of material density images expressing the density of the second reference material,
• each virtual monochromatic X-ray image of the first plurality of material density images and each virtual monochromatic X-ray image of the second plurality of material density images being images generated based on a virtual monochromatic X-ray image inferred by the first trained neural network,
• each material density image of the third plurality of material density images and each material density image of the fourth plurality of material density images being images generated by dual energy CT data, and
• the second neural network performing learning using the first plurality of material density images, second plurality of material density images, third plurality of material density images, and fourth plurality of material density images, such that each material density image of the first plurality of material density images and each material density image of the second plurality of material density images is used as input for the second neural network and each material density image of the third plurality of material density images and each material density image of the fourth plurality of material density images is used as output from the second neural network.