Granted Patent B2

US 12,266,116 · App. 17/643,409 · Granted Apr 1, 2025

Inventors: Yang Lyu (Shanghai, CN); Chen Xi (Shanghai, CN)

Assignee: SHANGHAI UNITED IMAGING HEALTHCARE CO., LTD.

G06T7/246A61B6/5235

Abstract

Claims (68)

1. A system, comprising:

at least one storage device storing executable instructions for model training; and

at least one processor in communication with the at least one storage device, wherein when executing the executable instructions, the at least one processor is configured to cause the system to perform operations including:

obtaining training samples, the training samples including at least one sample first image generated based on a first tracer, at least one reference first image each of which corresponds to one of the at least one sample first image and has a higher image quality than the corresponding sample first image, at least one sample second image generated based on a second tracer different from the first tracer, and at least one reference second image each of which corresponds to one of the at least one sample second images and has a higher image quality than the corresponding sample second image; and

generating a trained image processing model by training a preliminary model using the training samples.

2. The system of claim 1 , wherein obtaining the training samples includes:

obtaining one of the at least one reference first image generated based on first imaging data acquired by an imaging device according to the first tracer;

generating the corresponding sample first image by down-sampling the first imaging data;

obtaining one of the at least one reference second image generated based on second imaging data acquired by the imaging device according to the second tracer; and

generating the corresponding sample second image by down-sampling the second imaging data.

3. The system of claim 1 , wherein a first count of the at least one sample first image or the at least one reference first image is greater than a second count of the at least one sample second image or the at least one reference second image.

4. The system of claim 1 , wherein generating the trained image processing model by training the preliminary model using the training samples includes:

for each of the at least one sample first image, obtaining a first preliminary weight;

for each of the at least one sample second image, obtaining a second preliminary weight;

determining a loss function based on the at least one first preliminary weight and the at least one second preliminary weight; and

generating the trained image processing model by training the preliminary model based on the loss function.

5. The system of claim 4 , wherein the loss function includes a first term at least related to the at least one first preliminary weight and a second term at least related to the at least one second preliminary weight.

6. The system of claim 4 , wherein the at least one second preliminary weight is greater than the at least one first preliminary weight.

7. The system of claim 4 , wherein the training process includes a plurality of iterations, at least one iteration of the plurality of iterations including:

determining at least one first prediction output by inputting the at least one sample first image into a first intermediate model, the first intermediate model being the preliminary model in the first iteration of the plurality of iterations, or an updated model generated in a previous iteration;

determining at least one second prediction output by inputting the at least one sample second image into the first intermediate model;

obtaining at least one first current weight corresponding to the at least one sample first image, and at least one second current weight corresponding to the at least one sample second image, the at least one first current weight being the at least one first preliminary weight in the first iteration of the plurality of iterations, or at least one updated weight corresponding to the at least one first preliminary weight in the previous iteration, the at least one second current weight being the at least one second preliminary weight in the first iteration of the plurality of iterations, or at least one updated weight corresponding to the at least one second preliminary weight in the previous iteration;

determining a value of the loss function based on the at least one first current weight, the at least one second current weight, the at least one first prediction output, the at least one reference first image, the at least one second prediction output, and the at least one reference second image; and

updating at least one parameter of the first intermediate model based on the value of the loss function.

8. The system of claim 7 , wherein the at least one iteration of the plurality of iterations further includes:

determining a current image difference between the at least one reference second image and the at least one second prediction output; and

updating the at least one first current weight and the at least one second current weight based on the current image difference.

9. The system of claim 8 , wherein determining the current image difference between the at least one reference second image and the at least one second prediction output includes:

determining pixel differences between a plurality of pixels of each of the at least one reference second image and a plurality of pixels of each of the at least one second prediction output; and

determining the current image difference based on the pixel differences and the at least one second current weight.

10. The system of claim 8 , wherein updating the at least one first current weight and the at least one second current weight based on the current image difference includes:

obtaining an initial factor;

determining a target factor by comparing the current image difference of the current iteration and a previous image difference that corresponds to the current image difference and is determined in the previous iteration; and

updating the at least one first current weight and the at least one second current weight based on the comparison.

11. The system of claim 10 , wherein updating the at least one first current weight and the at least one second current weight based on the comparison includes:

in response to determining that the current image difference is greater than the previous image difference,

decreasing the at least one first current weight; and

increasing the at least one second current weight.

12. The system of claim 10 , wherein updating the at least one first current weight and the at least one second current weight based on the comparison includes:

in response to determining that the current image difference is less than the previous image difference,

increasing the at least one first current weight; and

decreasing the at least one second current weight.

13. The system of claim 1 , wherein generating the trained image processing model by training the preliminary model using the training samples includes:

generating a second intermediate model by training the preliminary model using the at least one sample first image and the at least one reference first image; and

generating the trained image processing model by training the second intermediate model using the at least one sample second image and the at least one reference second image.

14. The system of claim 13 , wherein generating the trained image processing model by training the second intermediate model using the at least one sample second image and the at least one reference second image includes:

generating the trained image processing model by iteratively updating parameter values of preset layers in the second intermediate model based on the at least one sample second image and the at least one reference second image.

15. The system of claim 14 , wherein there are a threshold count of layers between the preset layers and an output layer of the second intermediate model, and the threshold count of layers is less than 3.

16. The system of claim 1 , wherein

the trained image processing model is a trained Feedback Convolutional Neural Network (FB-CNN) including a plurality of sequentially connected subnets, and

an input of the FB-CNN is connected to an output of each of the plurality of subnets.

17. The system of claim 16 , wherein

each of the plurality of subnets includes at least one convolution block, at least one deconvolution block, and a feedback block (FB-block), an output of the FB-block in the subnet being inputted into a next subnet connected to the subnet,

the FB-block includes a plurality of convolutional layers and deconvolutional layers, a portion of the plurality of convolutional layers and deconvolutional layers being interleaved, and

different layers in at least part of the plurality of convolutional layers and deconvolutional layers are connected to each other.

18. The system of claim 1 , wherein

the at least one sample first image and the at least one sample second image serve as inputs for training the preliminary model, and

the at least one reference first image and the at least one reference second image serve as reference outputs for training the preliminary model.

19. A system, comprising:

at least one storage device storing executable instructions for image processing; and

at least one processor in communication with the at least one storage device, wherein when executing the executable instructions, the at least one processor is configured to cause the system to perform operations including:

obtaining an initial image of a subject acquired by an imaging device according to a second tracer;

obtaining a trained image processing model; and

generating a target image by inputting the initial image into the trained image processing model, the target image having a higher image quality than the initial image;

wherein the trained image processing model is provided by training a preliminary model using training samples, the training samples including at least one sample first image generated based on a first tracer, at least one reference first image each of which corresponds to one of the at least one sample first image and has a higher image quality than the corresponding sample first image, at least one sample second image generated based on the second tracer different from the first tracer, and at least one reference second image each of which corresponds to one of the at least one sample second image and has a higher image quality than the corresponding sample second image.

20. A method for model training, implemented on a computing device having at least one processor and at least one storage device, the method comprising:

obtaining training samples, the training samples including at least one sample first image generated based on a first tracer, at least one reference first image each of which corresponds to one of the at least one sample first image and has a higher image quality than the corresponding sample first image, at least one sample second image generated based on a second tracer different from the first tracer, and at least one reference second image each of which corresponds to one of the at least one sample second images and has a higher image quality than the corresponding sample second image; and

generating a trained image processing model by training a preliminary model using the training samples.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (24)