Ultrasound system with a neural network for producing images from undersampled ultrasound data

A method of generating medical images from sparsely sampled imaging data, the method comprising:

• receiving undersampled imaging data corresponding to images acquired during cardiac imaging;
• coupling the undersampled imaging data to a neural network to produce modified imaging data representative of sufficiently-sampled imaging data, the neural network is trained by:
• providing a first data set comprising sufficiently-sampled imaging data corresponding to previously-acquired cardiac images;
• producing a second data set with missing data by reducing spatial and temporal sampling of the first data set to mimic undersampling caused by low frame rate acquisition as applied to the cardiac imaging over at least one cardiac cycle;
• coupling the second data set to one or more neural networks including the neural network to generate new data estimated to represent the missing data;
• producing a third data set by modifying the second data set to include the new data;
• classifying, by the one or more neural networks, the third data set as either real or fake;
• and
• adjusting activation rules for one or more nodes of the one or more neural networks based on an accuracy of the classification;
• and
• generating one or more medical images based on the modified data.

A non-transitory computer-readable medium comprising executable instructions, which when executed cause a processor of a medical imaging system to perform the steps of:

• receiving undersampled imaging data corresponding to images acquired during cardiac imaging;
• coupling the undersampled imaging data to a neural network to produce modified imaging data representative of sufficiently-sampled imaging data, the neural network is trained by:
• providing a first data set comprising sufficiently-sampled imaging data corresponding to previously-acquired cardiac images;
• producing a second data set with missing data by reducing spatial and temporal sampling of the first data set to mimic undersampling caused by low frame rate acquisition as applied to the cardiac imaging over at least one cardiac cycle;
• coupling the second data set to one or more neural networks including the neural network to generate new data estimated to represent the missing data;
• producing a third data set by modifying the second data set to include the new data;
• classifying, by the one or more neural networks, the third data set as either real or fake;
• and
• adjusting activation rules for one or more nodes of the one or more neural networks based on an accuracy of the classification;
• and
• generating one or more medical images based on the modified data.