Methods and systems for processing an image

1 . A method being executed by a computing device including at least one processor and storage medium, comprising:

obtaining a first image;

determining a second image by optimizing the first image, wherein a quality of the second image is better than a quality of the first image, the first image is optimized using an optimization model, the optimization model includes a trained second machine learning model, and the optimization model is selected from a plurality of trained second machine learning models;

the plurality of trained second machine learning models correspond to different image resolution ranges, and the selection is based on a resolution of the first image; and

obtaining a third image by inputting the second image and a K-space dataset determined based on the second image into a fidelity model, wherein the fidelity model includes a trained machine learning model.

2 . The method of claim 1 , wherein a training process of the fidelity model includes:

obtaining a plurality of training samples, each of the plurality of training samples including one or more sample images and one or more sample K-space datasets, the one or more sample images including a sample second image determined by optimizing a sample first image and a gold-standard sample image corresponding to the sample second image, wherein the gold-standard sample image has a higher signal-to-noise ratio or a higher resolution than the sample first image, the one or more sample K-space datasets including a sample second K-space dataset corresponding to the sample second image and a sample reference K-space dataset, the sample reference K-space dataset being determined based on the sample first image and the sample second image or determined based on the sample second K-space dataset and a sample first K-space dataset corresponding to the sample first image;

training a preliminary fidelity model based on the plurality of training samples; and

in response to that a preset condition is met, obtaining the fidelity model.

3 . The method of claim 2 , wherein at least one loss function of the fidelity model includes a loss function in an image domain and a loss function in a K-space data domain;

the training process includes training the preliminary fidelity model based on the plurality of training samples by collectively using the loss function in the image domain and the loss function in the K-space data domain.

4 . The method of claim 3 , wherein

the loss function in the image domain is obtained based on the sample second image and the gold-standard sample image.

5 . The method of claim 3 , wherein

the loss function in the K-space data domain is obtained based on the sample second K-space dataset and the sample reference K-space dataset.

6 . The method of claim 2 , wherein the determining the sample reference K-space dataset based on a first K-space dataset corresponding to the sample first image and the second K-space dataset corresponding to the sample second image includes:

designating K-space center data of the sample first K-space dataset as K-space center data of the sample reference K-space dataset.

7 . The method of claim 6 , wherein determining the sample reference K-space dataset further comprises:

performing a fusion operation on K-space data of a surrounding region in the sample second K-space dataset and K-space data of a surrounding region in the sample first K-space dataset based on weights;

wherein a weight of the surrounding region from the sample second K-space dataset is greater than a weight of the surrounding region from the sample first K-space dataset.

8 . The method of claim 2 , wherein the sample reference K-space dataset is determined according to operations including:

determining a fusion level, the fusion level representing a proportional relationship between the sample first image and the sample second image or the fusion level representing a proportional relationship between the sample first K-space dataset and the sample second K-space dataset; and

determining, based on the fusion level, the sample reference K-space dataset.

9 . The method of claim 8 , further comprising:

determining the fusion level based on at least one of a scanning type, a scanning site, or a scanning protocol.

10 . The method of claim 8 , wherein the fusion level is spatially varied across the K-space based on a distance of a region from a K-space center.

11 . A system, comprising:

at least one storage device storing a set of instructions; and

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

obtaining a first image;

determining a second image by optimizing the first image, wherein a quality of the second image is better than a quality of the first image, the first image is optimized using an optimization model, the optimization model includes a trained second machine learning model, and the optimization model is selected from a plurality of trained second machine learning models; the plurality of trained second machine learning models correspond to different image resolution ranges, and the selection is based on a resolution of the first image; and

obtaining a third image by inputting the second image and a K-space dataset determined based on the second image into a fidelity model, wherein the fidelity model include a trained machine learning model.

12 . The system of claim 11 , wherein a training process of the fidelity model includes:

obtaining a plurality of training samples, each of the plurality of training samples including one or more sample images and one or more sample K-space datasets, the one or more sample images including a sample second image determined by optimizing a sample first image and a gold-standard sample image corresponding to the sample second image, wherein the gold-standard sample image has a higher signal-to-noise ratio or a higher resolution than the sample first image, the one or more sample K-space datasets including a sample second K-space dataset corresponding to the sample second image and a sample reference K-space dataset, the sample reference K-space dataset being determined based on the sample first image and the sample second image or determined based on the sample second K-space dataset and a sample first K-space dataset corresponding to the sample first image;

training a preliminary fidelity model based on the plurality of training samples; and

in response to that a preset condition is met, obtaining the fidelity model.

13 . The system of claim 12 , wherein at least one loss function of the fidelity model includes a loss function in an image domain and a loss function in a K-space data domain;

the training process includes training the preliminary fidelity model based on the plurality of training samples by collectively using the loss function in the image domain and the loss function in the K-space data domain.

14 . A non-transitory computer readable medium storing instructions, the instructions, when executed by at least one processor, causing the at least one processor to implement a method comprising:

obtaining a first image;

determining a second image by optimizing the first image, wherein a quality of the second image is better than a quality of the first image, the first image is optimized using an optimization model, the optimization model includes a trained second machine learning model;

the trained second machine learning model is trained based on sample noise images and one or more auxiliary parameters include at least one of a color, a depth, or a normal vector derived from each of the sample noise images; and

obtaining a third image by inputting the second image and a K-space dataset determined based on the second image into a fidelity model, wherein the fidelity model includes a trained machine learning model.