Apparatus and methods for unsupervised image denoising using double over-parameterization
A method, apparatus, and non-transitory computer-readable storage medium for image denoising whereby a deep image prior (DIP) neural network is trained to produce a denoised image by inputting the second medical image to the DIP neural network and combining a converging noise and an output of the DIP network during the training such that the converging noise combined with the output of the DIP network approximates the first medical image at the end of the training, wherein the output of the DIP network represents the denoised image.
1 . A method for denoising an image, the method comprising:
receiving a first medical image including a first image of an anatomical structure;
receiving a second medical image including a second image of the anatomical structure; and
training at least one deep image prior (DIP) neural network to produce a denoised image by inputting the second medical image to the at least one DIP neural network and combining a converging noise and an output of the at least one DIP network during the training such that the converging noise combined with the output of the at least one DIP network approximates the first medical image at the end of the training, wherein the output of the DIP network represents the denoised image,
wherein the step of training the at least one DIP neural network comprises
initializing first and second noise vectors;
training the at least one DIP neural network to produce the denoised image by training the first and second noise vectors to be equal to values for which a convolution-based function based on the first and second noise vectors converges to a noise of the first medical image; and
training the at least one DIP neural network to approximate the first medical image minus the convolution-based function, and
wherein the step of training the at least one DIP neural network further comprises training a plurality of DIP neural networks as the at least one DIP neural network using respectively different parameters.
2 . The method according to claim 1 , wherein the step of training the at least one DIP neural network comprises using a double over-parameterized training process on the converging noise.
3 . The method according to claim 1 , wherein the first medical image is a Position Emission Tomography (PET) image of a subject.
4 . The method according to claim 3 , wherein the second medical image is a Computed Tomography (CT) image of the subject registered to the PET image.
5 . The method according to claim 3 , wherein the second medical image is a Magnetic Resonance Imaging (MRI) image of the subject registered to the PET image.
6 . The method according to claim 1 , wherein the first medical image is single-photon emission computerized tomography (SPECT) image of a subject.
7 . The method according to claim 6 , wherein the second medical image is a Computed Tomography (CT) image of the subject registered to the SPECT image.
8 . The method according to claim 6 , wherein the second medical image is a Magnetic Resonance Imaging (MRI) image of the subject registered to the SPECT image.
9 . The method according to claim 1 , wherein the first medical image is an ungated cardiac Computed Tomography (CT) image of a subject.
10 . The method according to claim 9 , wherein the second medical image is a gated CT image of the subject registered to the ungated cardiac CT image.
11 . A medical image processing apparatus, comprising:
processing circuitry configured to:
receive a first medical image including a first image of an anatomical structure;
receive a second medical image including a second image of the anatomical structure; and
train at least one deep image prior (DIP) neural network to produce a denoised image by inputting the second medical image to the at least one DIP neural network and combining a converging noise and an output of the at least one DIP network during the training such that the converging noise combined with the output of the at least one DIP network approximates the first medical image at the end of the training, wherein the output of the at least one DIP network represents the denoised image,
wherein, in training the at least one DIP neural network, the processing circuitry is further configured to
initialize first and second noise vectors;
train the at least one DIP neural network to produce the denoised image by training the first and second noise vectors to be equal to values for which a convolution-based function based on the first and second noise vectors converges to a noise of the first medical image; and
train the at least one DIP neural network to approximate the first medical image minus the convolution-based function, and
wherein the processing circuitry is further configured to train a plurality of DIP neural networks as the at least one DIP neural network using respectively different parameters.
12 . The apparatus according to claim 11 , wherein the processing circuitry configured to train the at least one DIP neural network is further configured to use a double over-parameterized training process on the converging noise.
13 . A non-transitory computer-readable storage medium storing computer-readable instructions that, when executed by a computer, cause the computer to perform the steps of:
receiving a first medical image including a first image of an anatomical structure;
receiving a second medical image including a second image of the anatomical structure; and
training at least one deep image prior (DIP) neural network to produce a denoised image by inputting the second medical image to the at least one DIP neural network and combining a converging noise and an output of the at least one DIP network during the training such that the converging noise combined with the output of the at least one DIP network approximates the first medical image at the end of the training, wherein the output of the at least one DIP network represents the denoised image,
wherein the step of training the at least one DIP neural network comprises
initializing first and second noise vectors;
training the at least one DIP neural network to produce the denoised image by training the first and second noise vectors to be equal to values for which a convolution-based function based on the first and second noise vectors converges to a noise of the first medical image; and
training the at least one DIP neural network to approximate the first medical image minus the convolution-based function, and
wherein the step of training the at least one DIP neural network further comprises training a plurality of DIP neural networks as the at least one DIP neural network using respectively different parameters.
14 . The non-transitory computer-readable storage medium according to claim 13 , wherein the step of training the at least one DIP neural network comprises using a double over-parameterized training process on the converging noise.