Virtual microwave phantom generation
View Patent ↗A method for generating virtual microwave (MW) phantoms at a specific MW frequency or multiple MW frequencies, comprising generating virtual MW phantoms using a generative neural network (generator) and identifying the authenticity of the virtual MW phantoms using a discriminative neural network (discriminator).
1. A method for generating virtual microwave (MW) phantoms at a specific MW frequency, comprising:
generating virtual MW phantoms using a generative neural network (generator);
identifying the authenticity of the virtual MW phantoms using a discriminative neural network (discriminator);
wherein the generator is configured to generate virtual permittivity images or conductivity images at the specific MW frequency;
wherein the discriminator is configured to evaluate input permittivity images or conductivity images at the specific MW frequency;
selecting reconstructed MRI images from a searchable database of a plurality of reconstructed MRI images based on patient characteristics; and
converting the selected reconstructed MRI images to the input permittivity images or conductivity images at the specific MW frequency.
2. The method of claim 1 , further comprising training the generator and the discriminator in an alternative manner until a stopping criterion is reached wherein the virtual MW phantoms generated by the generator are realistic.
3. The method of claim 2 , wherein training the generator and the discriminator occurs using at least one of a plurality of computational processors or graphics processing units operating in parallel.
4. The method of claim 2 , further comprising generating electric fields from the virtual realistic MW phantoms using computational electromagnetic simulations.
5. The method of claim 4 , further comprising training a neural network using the generated electric fields as input and the virtual realistic MW phantoms as output.
6. The method of claim 5 , further comprising:
receiving collected electric fields; and
inputting the received collected electric fields to the neural network to reconstruct at least one permittivity image or conductivity image at the specific MW frequency.
7. A method for generating virtual microwave (MW) phantoms at a specific MW frequency, comprising:
generating virtual MW phantoms using a generative neural network (generator);
identifying the authenticity of the virtual MW phantoms using a discriminative neural network (discriminator);
wherein the generator comprises a plurality of transposed convolutional layers, batch normalization layers, and non-linear activation functions and wherein the discriminator comprises a plurality of convolutional layers, batch normalization layers, and non-linear activation functions.
8. The method of claim 7 , wherein at least a subset of the transposed convolutional layers are replaced by fully-connected layers.
9. A method for generating virtual microwave (MW) phantoms at multiple MW frequencies, comprising:
generating virtual MW phantoms using a generative neural network (generator);
identifying the authenticity of the virtual MW phantoms using a discriminative neural network (discriminator);
wherein the generator is configured to simultaneously generate a set of wideband MW images for each virtual MW phantom of the MW phantoms at the multiple MW frequencies; wherein the set of MW images comprises:
an image showing the distribution of the permittivity (ε ∞ ) at infinite high frequency;
an image showing the distribution of the permittivity (ε s ) at static frequency; and
an image showing the static conductivity (σ s ) at static frequency.
10. The method of claim 9 , wherein the set of wideband MW images further comprises: an image showing the distribution of relaxation parameter a; and an image showing the distribution of principal relaxation time (τ) of dipole rotation.
11. The method of claim 9 , wherein the discriminator is configured to evaluate input permittivity images or conductivity images at the multiple MW frequencies.
12. The method of claim 9 , wherein the virtual MW phantoms are virtual 3-D MW breast phantoms.
13. A method for generating virtual microwave (MW) phantoms, comprising:
generating virtual MRI phantoms using a generative neural network (generator);
identifying the authenticity of the virtual MRI phantoms using a discriminative neural network discriminator
converting the virtual MRI phantoms to virtual MW phantoms;
training the generator and the discriminator in an alternative manner until a stopping criterion is reached wherein the virtual MRI phantoms generated by the generator are realistic and hence conversion to virtual MW phantoms is realistic;
generating electric fields from the virtual realistic MW phantoms using computational electromagnetic simulations; and
training a neural network using the generated electric fields as input and the virtual realistic MW phantoms as output.
14. The method of claim 13 , further comprising:
receiving collected electric fields from at least one data acquisition site; and
inputting the received collected electric fields to the neural network to reconstruct at least one of (1) a permittivity image and conductivity image at a specific MW frequency or (2) a set of wideband MW images showing the distribution of permittivity and conductivity at a plurality of MW frequencies.
15. The method of claim 13 , further comprising:
selecting reconstructed MRI images from a searchable database of a plurality of reconstructed MRI images based on patient characteristics; and
wherein the discriminator is configured to evaluate the selected reconstructed MRI images.