Method for high-resolution image reconstruction

1 . A computer-implemented method comprising:

obtaining a plurality of images from a plurality of image scanning orientations;

performing a rigid registration to the plurality of images to obtain a transformation matrix to normalize the plurality of images from respective image spaces of the plurality of images to a normalized image space, and to obtain normalized images, wherein each normalized image comprises a plurality of voxels;

training a machine learning model comprising an implicit representation of a high-resolution image using the normalized images to obtain a trained machine learning model, wherein the high-resolution image comprises more voxels than the voxels in the normalized images; and

generating the high-resolution image based on the trained machine learning model;

wherein generating the high-resolution image based on the trained machine learning model comprises:

generating a dense grid of the normalized image space, wherein the dense grid is represented by voxels and each voxel is represented by a positional coordinate in the normalized image space;

inputting positional coordinates of voxels of the high-resolution image into an implicit representation model;

receiving intensity values from the implicit representation model of the trained machine learning model based on the positional coordinates of the voxels of the high-resolution image; and

generating the high-resolution image based on the intensity values.

2 . The computer-implemented method of claim 1 , further comprising:

performing Fourier feature mapping by mapping positional coordinates of the voxels to a higher dimensional space.

3 . The computer-implemented method of claim 1 , wherein the plurality of images are arranged in image slices in orthogonal orientations, and wherein the orthogonal orientations include at least one of a coronal orientation, an axial orientation, or a sagittal orientation.

4 . The computer-implemented method of claim 1 , wherein the plurality of images are a plurality of anisotropic 2D images, and the high-resolution image is a 2D high-resolution image.

5 . The computer-implemented method of claim 1 , wherein the plurality of images are a plurality of anisotropic 2D images, and the high-resolution image is a 3D high-resolution image.

6 . The computer-implemented method of claim 1 , wherein the machine learning model is optimized by minimizing errors between intensity values outputted by the machine learning model and intensity values of the normalized images.

7 . The computer-implemented method of claim 1 , wherein each image is selected from a group consisting of a magnetic resonance image, a computed tomography scan image, an optical image, and an electron scanning microscope image.

8 . The computer-implemented method of claim 1 , wherein the machine learning model is a fully connected multi-layer perceptron (MLP) network.

9 . A computer-implemented method comprising:

obtaining a plurality of images from a plurality of image scanning orientations;

performing a rigid registration to the plurality of images to obtain a transformation matrix to normalize the plurality of images from respective image spaces of the plurality of images to a normalized image space, and to obtain normalized images, wherein each normalized image comprises a plurality of voxels;

training a machine learning model comprising an implicit representation of a high-resolution image using the normalized images to obtain a trained machine learning model, wherein the high-resolution image comprises more voxels than the voxels in the normalized images; and

generating the high-resolution image based on the trained machine learning model;

wherein the machine learning model is a fully connected multi-layer perceptron (MLP) network; and

wherein the fully connected MLP network comprises eighteen fully-connected layers, wherein a number of neurons of first, seventh, and thirteenth fully-connected layers of the fully-connected MLP network is different from a number of neurons of other fully-connected layers of the fully-connected MLP network, and wherein the other fully-connected layers of the fully-connected MLP network have at least 256 neurons per layer.

10 . A non-transitory storage medium of a computing system storing instructions, wherein when the instructions are executed by one or more processors of the computing system, the computing system performs a method comprising:

obtaining a plurality of images from a plurality of image scanning orientations;

performing a rigid registration to the plurality of images to obtain a transformation matrix to normalize the plurality of images from respective image spaces of the plurality of images to a normalized image space, and to obtain normalized images, wherein each normalized image comprises a plurality of voxels;

training a machine learning model comprising an implicit representation of a high-resolution image using the normalized images to obtain a trained machine learning model, wherein the high-resolution image comprises more voxels than the voxels in the normalized images; and

generating the high-resolution image based on the trained machine learning model;

wherein generating the high-resolution image based on the trained machine learning model comprises:

generating a dense grid of the normalized image space, wherein the dense grid is represented by voxels and each voxel is represented by a positional coordinate in the normalized image space; inputting positional coordinates of voxels of the high-resolution image into an implicit representation model;

receiving intensity values from the implicit representation model of the trained machine learning model based on the positional coordinates of the voxels of the high-resolution image; and

generating the high-resolution image based on the intensity values.

11 . The non-transitory storage medium of claim 10 , wherein the method further comprises:

performing Fourier feature mapping by mapping positional coordinates of the voxels to a higher dimensional space.

12 . The non-transitory storage medium of claim 10 , wherein the plurality of images are arranged in image slices in orthogonal orientations, and wherein the orthogonal orientations include at least one of a coronal orientation, an axial orientation, or a sagittal orientation.

13 . The non-transitory storage medium of claim 10 , wherein the plurality of images are a plurality of anisotropic 2D images, and the high-resolution image is a 2D high-resolution image.

14 . The non-transitory storage medium of claim 10 , wherein the plurality of images are a plurality of anisotropic 2D images, and the high-resolution image is a 3D high-resolution image.

15 . The non-transitory storage medium of claim 10 , wherein the machine learning model is optimized by minimizing errors between intensity values outputted by the machine learning model and intensity values of the normalized images.

16 . The non-transitory storage medium of claim 10 , wherein each image is selected from a group consisting of a magnetic resonance image, a computed tomography scan image, an optical image, and an electron scanning microscope image.

17 . The non-transitory storage medium of claim 10 , wherein the machine learning model is a fully connected multi-layer perceptron (MLP) network.

18 . The non-transitory storage medium of claim 17 , wherein the fully connected MLP network comprises eighteen fully-connected layers, wherein a number of neurons of first, seventh, and thirteenth fully-connected layers of the fully-connected MLP network is different from a number of neurons of other fully-connected layers of the fully-connected MLP network, and wherein the other fully-connected layers of the fully-connected MLP network have at least 256 neurons per layer.