Network training and architecture for medical imaging

A method comprising:

• acquiring a plurality of non-attenuation-corrected volumes, each of the non-attenuation-corrected volumes based on a respective one of a plurality of sets of two-dimensional emission data;
• acquiring a plurality of attenuation coefficient maps, each of the plurality of attenuation coefficient maps corresponding to a respective one of the plurality of sets of two-dimensional emission data;
• training a convolutional network to generate an attenuation coefficient map from an input image volume, the training based on the plurality of non-attenuation-corrected volumes and respective ones of the plurality of attenuation coefficient maps;
• and
• outputting trained kernels of the trained convolutional network to an emission imaging system.

A system comprising:

• a storage device;
• a processor to execute processor-executable process steps stored on the storage device to cause the system to:
• acquire a plurality of non-attenuation-corrected volumes, each of the non-attenuation-corrected volumes based on a respective one of a plurality of sets of two-dimensional emission data;
• train a first convolutional network to generate an image volume from an input set of two-dimensional emission data, the training based on the plurality of non-attenuation-corrected volumes and respective ones of the plurality of sets of two-dimensional image data;
• acquire a plurality of attenuation coefficient maps, each of the plurality of attenuation coefficient maps corresponding to a respective one of a second plurality of sets of two-dimensional emission data;
• train a second convolutional network to generate an attenuation coefficient map from an input image volume, the training based on a plurality of volumes generated by the first convolutional network in response to input to the first convolutional network of each of the second plurality of sets of two-dimensional emission data, and on the plurality of attenuation coefficient maps;
• and
• output first trained kernels of the first trained convolutional network and second trained kernels of the second trained convolutional network to an emission imaging system.