Trained model generation program, image generation program, trained model generation device, image generation device, trained model generation method, and image generation method
A trained model generation program causes a computer to implement a learning execution function of inputting first input image data representing a first input image generated by a first reconstruction method using compressed sensing and second input image data representing a second input image generated by a second reconstruction method different from the first reconstruction method to a machine learning device to execute machine learning, the second reconstruction method being an analytical reconstruction method, and causing the machine learning device to generate a trained model, and a trained model acquisition function of acquiring trained model data indicating the trained model. Then, input image data representing an input image is input to the trained model to generate a reconstructed image with improved image quality.
1. A non-transitory computer-readable storage medium storing a trained model generation program causing a computer to implement:
a learning execution function of:
inputting first input image data representing a first input image generated by a first reconstruction method using compressed sensing and second input image data representing a second input image generated by a second reconstruction method different from the first reconstruction method to a machine learning device to execute machine learning, the second reconstruction method being an analytical reconstruction method, and
causing the machine learning device to generate a trained model; and
a trained model acquisition function of acquiring trained model data indicating the trained model,
wherein the learning execution function searches for a smoothing parameter of the first reconstruction method, a weight used in a convolutional neural network that is used as the trained model, and a bias used in the convolutional neural network which minimize a mean squared error to cause the machine learning device to generate the trained model.
2. The non-transitory computer-readable storage medium according to claim 1 , wherein the learning execution function searches for the smoothing parameter, the weight, and the bias which minimize the mean squared error represented by the following equation (1) to cause the machine learning device to generate the trained model:
[
Math
.
1
]
MSE
(
w
,
b
,
β
)
=
1
N
∑
i
=
1
N
x
i
-
CNN
(
w
,
b
)
[
y
i
z
i
(
β
)
]
2
(
1
)
where w is a vector in which weights used in the convolutional neural network are arranged in a row,
b is a vector in which biases used in the convolutional neural network are arranged in a row,
β is a smoothing parameter of the first reconstruction method,
x i is a vector in which values indicating densities represented by pixels of an output image are arranged in a row,
y i is a vector in which values indicating densities represented by pixels of the second input image are arranged in a row, and
z i is a vector in which values indicating densities represented by pixels of the first input image are arranged in a row.
3. The non-transitory computer-readable storage medium according to claim 1 , wherein the learning execution function:
inputs to the machine learning device at least the first input image data representing the first input image generated by the first reconstruction method using the compressed sensing configured with a first smoothing parameter and the first input image data representing the first input image generated by the first reconstruction method using the compressed sensing configured with a second smoothing parameter having a different value from the first smoothing parameter, and
causes the machine learning device to perform machine learning and generate the trained model.
4. The non-transitory computer-readable storage medium according to claim 1 , wherein the learning execution function:
inputs to the machine learning device output image data representing an output image generated by a third reconstruction method different from the first reconstruction method, in addition to the first input image data, and
causes the machine learning device to perform machine learning and generate the trained model.
5. An image generation program causing a computer to implement an image generation function of inputting input image data representing an input image to the trained model generated by the trained model generation program according to claim 1 to generate a reconstructed image.
6. A trained model generation device comprising:
a processor configured to:
input to a machine learning device first input image data representing a first input image generated by a first reconstruction method using compressed sensing and second input image data representing a second input image generated by a second reconstruction method different from the first reconstruction method, the second reconstruction method being an analytical reconstruction method;
cause the machine learning device to perform machine learning and generate a trained model; and
acquire trained model data indicating the trained model,
wherein the processor is configured to search for a smoothing parameter of the first reconstruction method, a weight used in a convolutional neural network that is used as the trained model, and a bias used in the convolutional neural network which minimize a mean squared error to cause the machine learning device to generate the trained model.
7. An image generation device comprising:
a processor configured to input input image data representing an input image to the trained model generated by the trained model generation device according to claim 6 to generate a reconstructed image.
8. A trained model generation method comprising:
inputting to a machine learning device first input image data representing a first input image generated by a first reconstruction method using compressed sensing and second input image data representing a second input image generated by a second reconstruction method different from the first reconstruction method, the second reconstruction method being an analytical reconstruction method;
causing the machine learning device to perform machine learning and generate a trained model; and
acquiring trained model data indicating the trained model,
wherein the method includes searching for a smoothing parameter of the first reconstruction method, a weight used in a convolutional neural network that is used as the trained model, and a bias used in the convolutional neural network which minimize a mean squared error to cause the machine learning device to generate the trained model.
9. An image generation method comprising inputting input image data representing an input image to the trained model generated by the trained model generation method according to claim 8 to generate a reconstructed image.
10. A non-transitory computer-readable storage medium storing a trained model generation program causing a computer implement:
a learning execution function of:
inputting first input image data including more than two first input images generated by a first reconstruction method using compressed sensing and second input image data including a second input image generated by a second reconstruction method different from the first reconstruction method to a machine learning device to execute machine learning, the second reconstruction method being an analytical reconstruction method, and
causing the machine learning device to generate a trained model; and
a trained model acquisition function of acquiring trained model data indicating the trained model,
wherein the more than two first input images are configured with different smoothing parameters from each other, and
wherein the learning execution function searches for a smoothing parameter of the first reconstruction method, a weight used in a convolutional neural network that is used as the trained model, and a bias used in the convolutional neural network which minimize a mean squared error to cause the machine learning device to generate the trained model.