Application of machine learning to iterative and multimodality image reconstruction
A method for machine learning based ultrasound image reconstruction can include receiving, at a reconstruction engine, imaging data; generating an initial estimate for a transmission image via a neural network trained (machine or self-learning) on paired transmission ultrasound and reflection ultrasound data; and performing image reconstruction using the initial estimate to generate transmission ultrasound images. The image reconstruction can generate higher quality transmission ultrasound when carried out by using the initial estimate as the starting point for iterative image reconstruction and using transmission data obtained via conventional transmission ultrasound frequencies (e.g. from 0.8 MHz to 1.5 MHz).
1. A system comprising:
a communication interface receiving imaging data from a scanner system; and
a reconstruction engine comprising:
at least one neural network that includes a GAN-based algorithm that receives at least a component of the imaging data from the communication interface;
a processor; and
storage media storing instructions for performing image reconstruction using an output of the at least one neural network when executed by the processor,
wherein the at least one neural network comprises two neural networks;
wherein the imaging data is reflection data and a reflection image;
wherein one of the two neural networks receives the reflection data and generates a set of synthetic transmission data from the reflection data; and
wherein the other one of the two neural networks receives the reflection image and generates an intermediate synthetic transmission image from the reflection image.
2. The system of claim 1 , wherein the instructions for performing image reconstruction, when executed by the processor, direct the processor to at least:
perform iterative methods on the set of synthetic transmission data and the intermediate synthetic transmission image until a desired accuracy is reached; and
upon reaching the desired accuracy, output a final synthetic transmission image.
3. The system of claim 1 , further comprising an image reconstruction selector application programming interface (API), wherein the image reconstruction selector API receives an image reconstruction algorithm for execution by the processor.
4. The system of claim 1 , further comprising a training data application programming interface (API), wherein the training data API receives training data or weights for use in training the neural network.
5. A system comprising:
a communication interface receiving imaging data from a scanner system; and
a reconstruction engine comprising:
at least one neural network that includes a GAN-based algorithm that receives at least a component of the imaging data from the communication interface;
a processor; and
storage media storing instructions for performing image reconstruction using an output of the at least one neural network when executed by the processor,
wherein the imaging data is high frequency transmission data and a reflection image; and
wherein the at least one neural network receives the reflection image and generates an intermediate synthetic transmission image from the reflection image.
6. The system of claim 5 , wherein the instructions for performing image reconstruction, when executed by the processor, direct the processor to at least:
perform iterative methods on the high frequency transmission data and the intermediate synthetic transmission image until a desired accuracy is reached; and
upon reaching the desired accuracy, output a final synthetic transmission image.
7. The system of claim 5 , further comprising an image reconstruction selector application programming interface (API), wherein the image reconstruction selector API receives an image reconstruction algorithm for execution by the processor.
8. The system of claim 5 , further comprising a training data application programming interface (API), wherein the training data API receives training data or weights for use in training the neural network.
9. A system comprising:
a communication interface receiving imaging data from a scanner system; and
a reconstruction engine comprising:
at least one neural network that includes a GAN-based algorithm that receives at least a component of the imaging data from the communication interface;
a processor; and
storage media storing instructions for performing image reconstruction using an output of the at least one neural network when executed by the processor,
wherein the reconstruction engine further comprises a discriminator and a repository of corresponding reflection and transmission data;
wherein a neural network of the at least one neural network receives reflection data and generates synthetic transmission data; and
wherein the discriminator compares transmission data corresponding to the reflection data to the synthetic transmission data and outputs a set of errors that are used to adjust weights in the GAN-based algorithm for the neural network.
10. The system of claim 9 , further comprising an image reconstruction selector application programming interface (API), wherein the image reconstruction selector API receives an image reconstruction algorithm for execution by the processor.
11. The system of claim 9 , further comprising a training data application programming interface (API), wherein the training data API receives training data or weights for use in training the neural network.
12. A system comprising:
a communication interface receiving imaging data from a scanner system; and
a reconstruction engine comprising:
at least one neural network that includes a GAN-based algorithm that receives at least a component of the imaging data from the communication interface;
a processor; and
storage media storing instructions for performing image reconstruction using an output of the at least one neural network when executed by the processor,
further comprising a discriminator and a repository of corresponding reflection and transmission images;
wherein a neural network of the at least one neural network receives a reflection image and generates a synthetic transmission image; and
wherein the discriminator compares a transmission image corresponding to the reflection image to the synthetic transmission image and outputs a set of errors that are used to adjust weights in the GAN-based algorithm.