US 12,361,520 B2
Systems and methods for deblurring and denoising medical images
Yikang Liu, Cambridge, MA (US); Zhang Chen, Brookline, MA (US); Xiao Chen, Lexington, MA (US); Shanhui Sun, Lexington, MA (US); and Terrence Chen, Lexington, MA (US)
Assigned to Shanghai United Imaging Intelligence Co., Ltd., Shanghai (CN)
Filed by Shanghai United Imaging Intelligence Co., Ltd., Shanghai (CN)
Filed on Nov. 17, 2022, as Appl. No. 17/989,205.
Prior Publication US 2024/0169486 A1, May 23, 2024
Int. Cl. G06T 5/73 (2024.01); G06T 5/50 (2006.01); G06T 5/70 (2024.01)
CPC G06T 5/50 (2013.01) [G06T 5/70 (2024.01); G06T 5/73 (2024.01); G06T 2207/10016 (2013.01); G06T 2207/10121 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30004 (2013.01)] 18 Claims
OG exemplary drawing
 
1. An apparatus, comprising:
at least one processor configured to:
obtain a sequence of input medical images;
process the sequence of input medical images through an artificial neural network (ANN), wherein the ANN is trained to reduce both blurriness and noise in the sequence of input medical images; and
generate, based on the processing, a sequence of output medical images that corresponds to the sequence of input medical images, wherein:
each of the output medical images is characterized by reduced blurriness and reduced noise compared to a corresponding one of the input medical images;
the ANN is trained using at least a first training dataset comprising medical images with synthetic noise and a second training dataset comprising medical images with real noise;
during the training of the ANN, the ANN is configured to predict respective deblurred and denoised medical images based on the medical images comprised in the first training dataset; and
parameters of the ANN are adjusted during the training based on at least a first loss designed to maintain continuity between consecutive medical images generated by the ANN, a second loss designed to maintain similarity of two or more patches inside a medical image generated by the ANN, and a third loss that indicates a difference between a deblurred and denoised medical image predicted by the ANN based on the first training dataset and a corresponding ground truth image.