Systems and methods for tailored image texture in iterative image reconstruction
Methods and systems are provided for reconstructing images with a tailored image texture. In one embodiment, a method comprises acquiring projection data, and reconstructing an image from the projection data with a desired image texture. In this way, iterative image reconstruction techniques may be used to substantially reduce image noise, thereby enabling a reduction in injected contrast and/or radiation dose, while preserving an image texture familiar from analytic image reconstruction techniques.
1. A method, comprising:
acquiring projection data;
reconstructing a first image from the projection data with a first image reconstruction method;
reconstructing a second image from the projection data with a second image reconstruction method;
blending the first image and the second image to generate a blended image with a desired image texture;
wherein the blended image includes an image texture of the first image and an image noise of the second image; and
wherein blending the first image and the second image to generate the image with the desired image texture comprises applying convolutions to the first image and the second image, and summing the images in image space; and
wherein the convolutions are calculated from a noise power spectrum of the first image.
2. The method of claim 1 , further comprising transforming the first image and the second image into frequency space, wherein blending the first image and the second image comprises summing a weighted combination of the first image and the second image in frequency space.
3. The method of claim 2 , further comprising multiplying a frequency representation of the first image with a weighted mask to generate a masked first image and multiplying a frequency representation of the second image with an inverted weighted mask to generate a masked second image, wherein summing a weighted combination of the first image and the second image in frequency space comprises summing the masked first image and the masked second image in frequency space.
4. The method of claim 3 , further comprising calculating the weighted mask from a noise power spectrum, wherein the inverted weighted mask comprises an inversion of the weighted mask.
5. The method of claim 4 , wherein calculating the weighted mask from the noise power spectrum comprises determining, with a trained neural network, a contribution of the first image at each spatial frequency to the image.
6. The method of claim 1 , wherein the first image reconstruction method comprises an iterative reconstruction method and wherein the second image reconstruction method comprises a filtered back projection (FBP) method.
7. The method of claim 1 , further comprising reconstructing, from the projection data with a filtered back projection (FBP) method, an FBP image, and inputting the FBP image into a trained neural network to generate the first image with a desired image texture.
8. The method of claim 1 , wherein the projection data is acquired by a computed tomography (CT) imaging system.
9. The method of claim 1 , wherein the first reconstruction method is an non-iterative reconstruction method and the second reconstruction method is an iterative reconstruction method.
10. The method of claim 7 , wherein the image texture is reinforced during a subset of iterations of the iterative reconstruction method.