US 12,148,073 B2
Deep encoder-decoder models for reconstructing biomedical images
Thomas C. Fuchs, New York, NY (US); Ida Häggström, New York, NY (US); and Charles Ross Schmidtlein, New York, NY (US)
Assigned to Memorial Sloan Kettering Cancer Center, New York, NY (US)
Appl. No. 17/040,416
Filed by MEMORIAL SLOAN KETTERING CANCER CENTER, New York, NY (US)
PCT Filed Mar. 22, 2019, PCT No. PCT/US2019/023733
§ 371(c)(1), (2) Date Sep. 22, 2020,
PCT Pub. No. WO2019/183584, PCT Pub. Date Sep. 26, 2019.
Claims priority of provisional application 62/647,190, filed on Mar. 23, 2018.
Claims priority of provisional application 62/734,038, filed on Sep. 20, 2018.
Prior Publication US 2021/0074036 A1, Mar. 11, 2021
Int. Cl. G06T 11/00 (2006.01); A61B 6/03 (2006.01); G06T 7/00 (2017.01); G06T 9/00 (2006.01)
CPC G06T 11/006 (2013.01) [A61B 6/032 (2013.01); A61B 6/037 (2013.01); G06T 7/0012 (2013.01); G06T 9/002 (2013.01); G06T 2207/10104 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30004 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method of reconstructing biomedical images, comprising:
identifying, by a computing system, a first projection dataset acquired from a first tomographic biomedical imaging scan in a first image modality;
applying, by the computing system, to the first projection dataset, an encoder-decoder model for reconstructing tomographic biomedical images from projection data, the encoder-decoder model comprising:
an encoder comprising a first series of transform layers having an input size, to generate a feature map of a first size using the first projection dataset; and
a decoder comprising a second series of transform layers and at least one up-sampler, the decoder having an output size different from the input size of the first series of transform layers of the encoder, the decoder to generate a first tomographic biomedical image of a second size using the feature map of the first size from the encoder, the second size larger than the first size via application the at least one up-sampler, the first tomographic biomedical image being in the first image modality;
wherein the encoder-decoder model is trained using (i) a second projection dataset acquired using a second tomographic biomedical imaging scan in a second image modality and (ii) a second tomographic biomedical image in the second image modality derived from the second projection dataset, the second image modality different from the first image modality of the first tomographic biomedical image, and
providing, by the computing system, the generated first tomographic biomedical image.