Neural network for image registration and image segmentation trained using a registration simulator
Apparatuses, systems, and techniques to perform registration among images. In at least one embodiment, one or more neural networks are trained to indicate registration of features in common among at least two images by generating a first correspondence by simulating a registration process of registering an image and applying the at least two images and the first correspondence to a neural network to derive a second correspondence of the features in common among the at least two images.
1 . A method, comprising:
using a processor comprising one or more circuits to;
compare a first image generated based, at least in part, on one or more pixel displacement values with a second image generated based, at least in part, on one or more other pixel displacement values; and
update a neural network based, at least in part, on a result of the comparison between the first image and the second image.
2 . The method of claim 1 , further comprising training the neural network to generate the first third image, wherein training the neural network comprises:
selecting a set of sampled transformation parameters from among a range of possible transformation parameters;
generating, from the set of sampled transformation parameters, a first displacement field including the one or more other pixel displacement values, the one or more other pixel displacement values indicating one or more displacements of pixels in a training image displaced according to the set of sampled transformation parameters;
transforming the training image into the second image according to the first displacement field;
generating, using the neural network, a second displacement field from the training image and the second image;
transforming the training image into the first image according to the second displacement field;
generating a loss function from a first difference between the first displacement field and the second displacement field; and
using the loss function to train the neural network.
3 . The method of claim 2 , wherein generating the loss function comprises generating the loss function from the first difference and from a second difference between the second image and the first image, wherein the loss function is a weighted sum of the first difference and the second difference.
4 . The method of claim 3 , wherein the first difference is a mean value over an image coordinate space of a norm of differences of elements between the first displacement field and the second displacement field, and wherein the second difference is a negative normalized cross-correlation between the second image and the first image.
5 . The method of claim 2 , wherein the range of possible transformation parameters to transform the training image into the second image comprises one or more of a range of rotation angles, a range of translations, a range of scale factors, and a range of elastic distortions, and wherein selecting the set of sampled transformation parameters comprises selecting a rotation angle within the range of rotation angles, selecting a translation within the range of translations, selecting a scale factor within the range of scale factors, and/or selecting an elastic distortion within the range of elastic distortions.
6 . The method of claim 2 , wherein the first training image is a two-dimensional or three-dimensional image of a biological environment.
7 . The method of claim 1 , further comprising training the neural network to generate the first image based, at least in part, on the one or more other pixel displacement values, wherein the training is performed in a supervised manner.
8 . A non-transitory computer-readable storage media storing executable instructions that, when executed by one or more processors of a computer system, cause the computer system to:
compare a first image generated based, at least in part, on one or more pixel displacement values with a second image generated based, at least in part, on one or more other pixel displacement values; and
update a neural network based, at least in part, on a result of the comparison between the first image and the second image.
9 . The non-transitory computer-readable storage media of claim 8 , wherein the executable instructions, when executed by the one or more processors of the computer system, further cause the computer system to train the neural network to generate the first image, wherein training the neural network comprises:
selecting a set of sampled transformation parameters from among a range of possible transformation parameters;
generating, from the set of sampled transformation parameters, a first displacement field including the other one or more pixel displacement values, the one or more other pixel displacement values indicating one or more displacements of pixels in a training image displaced according to the set of sampled transformation parameters;
transforming the training image into the second image according to the displacement field;
generating, using the neural network, a second displacement field from the training image and the second image;
transforming the training image into the first image according to the second displacement field;
generating a loss function from a first difference between the first displacement field and the second displacement field; and
using the loss function to train the neural network.
10 . The non-transitory computer-readable storage media of claim 9 , wherein generating the loss function comprises generating the loss function from the first difference and from a second difference between the second image and the first image, wherein the loss function is a weighted sum of the first difference and the second difference.
11 . The non-transitory computer-readable storage media of claim 10 , wherein the first difference is a mean value over an image coordinate space of a norm of differences of elements between the first displacement field and the second displacement field, and wherein the second difference is a negative normalized cross-correlation between the second image and the first image.
12 . The non-transitory computer-readable storage media of claim 9 , wherein the range of possible transformation parameters to transform the training image into the second image comprises one or more of a range of rotation angles, a range of translations, a range of scale factors, and a range of elastic distortions, and wherein selecting the set of sampled transformation parameters comprises selecting a rotation angle within the range of rotation angles, selecting a translation within the range of translations, selecting a scale factor within the range of scale factors, and/or selecting an elastic distortion within the range of elastic distortions.
13 . The non-transitory computer-readable storage media of claim 9 , wherein the training image is a two-dimensional or three-dimensional image of a biological environment.
14 . A processor, comprising:
one or more circuits to:
compare a first image generated based, at least in part, on one or more pixel displacement values with a second image generated based, at least in part, on one or more other pixel displacement values; and
update a neural network based, at least in part, on a result of the comparison between the first image and the second image.
15 . The processor of claim 14 , wherein the one or more circuits are further to train the neural network to generate the first image, wherein training the neural network comprises:
selecting a set of sampled transformation parameters from among a range of possible transformation parameters;
generating, from the set of sampled transformation parameters, a first displacement field including the one or more other pixel displacement values, the one or more other pixel displacement values indicating one or more displacements of pixels in a training image displaced according to the set of sampled transformation parameters;
transforming the training image into the second image according to the displacement field;
generating, using the neural network, a second displacement field from the training image and the second image;
transforming the training image into the first image according to the second displacement field;
generating a loss function from a first difference between the first displacement field and the second displacement field; and
using the loss function to train the neural network.
16 . The processor of claim 15 , wherein generating the loss function comprises generating the loss function from the first difference and from a second difference between the second image and the first image, wherein the loss function is a weighted sum of the first difference and the second difference.
17 . The processor of claim 16 , wherein the first difference is a mean value over an image coordinate space of a norm of differences of elements between the first displacement field and the second displacement field, and wherein the second difference is a negative normalized cross-correlation between the second image and the first image.
18 . The processor of claim 15 , wherein the range of possible transformation parameters to transform the training image into the second image comprises one or more of a range of rotation angles, a range of translations, a range of scale factors, and a range of elastic distortions, and wherein selecting the set of sampled transformation parameters comprises selecting a rotation angle within the range of rotation angles, selecting a translation within the range of translations, selecting a scale factor within the range of scale factors, and/or selecting an elastic distortion within the range of elastic distortions.
19 . The processor of claim 15 , wherein the training image is a two-dimensional or three-dimensional image of a biological environment.
20 . A computer system, comprising:
one or more processors and a computer-readable memory storing executable instructions that, when executed by the one or more processors, cause the computer system to:
compare a first image generated based, at least in part, on one or more pixel displacement values with a second image generated based, at least in part, on one or more other pixel displacement values; and
update one or more neural networks based, at least in part, on a result of the comparison between the first image and the second image.
21 . The computer system of claim 20 , wherein:
the executable instructions, when executed by the one or more processors, further cause the computer system to calculate parameters corresponding to the one or more neural networks, the calculation based, at least in part, on the comparison of the first image with the second image and providing a loss function for the one or more neural networks to optimize, the loss function including:
a first function of differences between a simulated displacement field and an output displacement field that the one or more neural networks outputs based on attempted registration of a training image and the second image; and
a second function of a first difference function of differences between a first displacement field including the one or more other pixel displacement values, the one or more other pixel displacement values indicating one or more displacements of pixels in the training image displaced according to a set of sampled transformation parameters, and a second displacement field, wherein the second displacement field is generated by the one or more neural networks from the output displacement field;
the computer system further comprises:
storage for the parameters;
storage for the set of sampled transformation parameters selected from among a range of possible transformation parameters; and
storage for the first displacement field; and
the parameters calculated based, at least in part, on a third function optimizing for the loss function.
22 . The computer system of claim 21 , wherein:
the loss function is a weighted sum of the first difference function and a second difference function of differences between the second image and the first image, the first image transformed from the training image according to the second displacement field; and
the first difference function is a mean value over an image coordinate space of a norm of differences of elements between the first displacement field and the second displacement field, and wherein the second difference function is a negative normalized cross-correlation between the second image and the first image.
23 . The computer system of claim 21 , wherein the range of possible transformation parameters to transform the training image into the second image comprises one or more of a range of rotation angles, a range of translations, a range of scale factors, and a range of elastic distortions, and wherein selecting the set of sampled transformation parameters comprises selecting an angle within the range of rotation angles, selecting a translation within the range of translations, selecting a scale factor within the range of scale factors, and/or selecting an elastic distortion within the range of elastic distortions.
24 . The computer system of claim 21 , wherein the training image is a two-dimensional or three-dimensional image of a biological environment.
25 . The computer system of claim 21 , further comprising:
a convolutional layer to modify a last neural network layer that provides a convolution of a last layer of the one or more neural networks and a predicted segmentation; and
a softmax activation layer to modify the predicted segmentation according to a softmax operation on the convolution.
26 . The computer system of claim 20 , wherein the executable instructions, when executed by the one or more processors, further cause the computer system to train the one or more neural networks to generate the first image, wherein training the one or more neural networks comprises:
selecting a set of sampled transformation parameters from among a range of possible transformation parameters;
generating, from the set of sampled transformation parameters, a first displacement field including the one or more other pixel displacement values, the one or more other pixel displacement values indicating one or more displacements of pixels in a training image displaced according to the set of sampled transformation parameters;
transforming the training image into the second image according to the first displacement field;
generating, using the one or more neural networks, a second displacement field from the training image and the second image;
transforming the training image into the first image according to the second displacement field;
generating a loss function from a first difference between the first displacement field and the second displacement field; and
using the loss function to train the one or more neural networks.
27 . The computer system of claim 26 , wherein:
generating the loss function comprises generating the loss function from the first difference and from a second difference between the second image and the first image, wherein the loss function is a weighted sum of the first difference and the second difference; and
the first difference is a mean value over an image coordinate space of a norm of differences of elements between the first displacement field and the second displacement field, and wherein the second difference is a negative normalized cross-correlation between the second image and the first image.
28 . The computer system of claim 26 , wherein the range of possible transformation parameters to transform the training image into the second image comprises one or more of a range of rotation angles, a range of translations, a range of scale factors, and a range of elastic distortions, and wherein selecting the set of sampled transformation parameters comprises selecting a rotation angle within the range of rotation angles, selecting a translation within the range of translations, selecting a scale factor within the range of scale factors, and/or selecting an elastic distortion within the range of elastic distortions.
29 . The computer system of claim 26 , wherein the training image is a two-dimensional or three-dimensional image of a biological environment.