US 12,032,048 B2
Machine learning for simultaneously optimizing an under-sampling pattern and a corresponding reconstruction model in compressive sensing
Mert R. Sabuncu, Ithaca, NY (US); and Cagla D Bahadir, Ithaca, NY (US)
Assigned to Cornell University, Ithaca, NY (US)
Appl. No. 17/416,281
Filed by Cornell University, Ithaca, NY (US)
PCT Filed Dec. 20, 2019, PCT No. PCT/US2019/067887
§ 371(c)(1), (2) Date Jun. 18, 2021,
PCT Pub. No. WO2020/132463, PCT Pub. Date Jun. 25, 2020.
Claims priority of provisional application 62/783,892, filed on Dec. 21, 2018.
Prior Publication US 2022/0075017 A1, Mar. 10, 2022
Int. Cl. A61B 5/055 (2006.01); A61B 5/00 (2006.01); G01R 33/56 (2006.01); G01R 33/561 (2006.01)
CPC G01R 33/561 (2013.01) [A61B 5/055 (2013.01); A61B 5/7203 (2013.01); A61B 5/7267 (2013.01); G01R 33/5608 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method of training a MRI image reconstruction system, comprising:
providing a set of input MRI images, wherein the input MRI images are all images of a same organ from different subjects and wherein each image from the set of input MRI images have a corresponding image quality and a corresponding k-space; and
processing the input MRI images using a neural network to collectively identify an under-sampling pattern and a reconstruction model, the under-sampling pattern and reconstruction model being specific to the same organ in the input MRI images;
producing new MRI images with the identified under-sampling pattern and reconstruction model, the new MRI images having an image quality that is substantially similar to the image quality of the input MRI images, from under-sampled MRI data to be captured using the identified under-sampling pattern, wherein the under-sampling pattern is configured to cause a k-space of the under-sampled MRI data to be under-sampled relative to the k-space of the input MRI images.