Method for providing a trainable function for determination of synthetic image data

1. A computer-implemented method for providing a trainable function for determination of synthetic image data, the method

comprising:

receiving first medical image data,

the first medical image data being based on a first medical imaging of an examination object;

receiving second medical image data,

the second medical image data being based on a second medical imaging of the examination object, the first and the second medical imaging differ in at least one of the imaging modality used or in an imaging protocol used, the first and the second medical image data are registered with one another;

determining the synthetic image data by applying the trainable function to the first medical image data;

determining a measure of similarity with a similarity function by comparison of the synthetic image data and the second medical image data, the determining the measure of similarity is based on an optimization method, and the optimization method comprises a maximization of the measure of similarity by at least one of a geometrical or photometric transformation of at least one of the synthetic image data or of the second medical image data, the optimization method including,

determining at least one first part area of the synthetic image data,

determining a plurality of second part areas of the second medical image data,

determining a plurality of part measures of similarity between the at least one first part area and the plurality of second part areas, and

determining the similarity function based on the plurality of part measures of similarity, wherein

the similarity function is based on a comparison of the first part area of the synthetic image data and a selected second part area of the second medical image data, and

the selected second part area corresponds to that area of the plurality of second part areas with the maximum part measure of similarity;

adjusting at least one parameter of the trainable function by optimization of the similarity function based on the measure of similarity; and

providing the trainable function.

2. The computer-implemented method as claimed in claim 1 ,

wherein at least one of the first or second medical imaging is at least one of a histopathology, an X-ray imaging, an angiography, a computed tomography, a magnetic resonance tomography, an ultrasound imaging, an electrocardiography, a positron emission tomography, a single photon emission computed tomography or an optical coherence tomography.

3. The computer-implemented method as claimed in claim 1 , wherein the at least one first part area and the plurality of second part areas have the same geometrical form.

4. The computer-implemented method as claimed in claim 1 , wherein the determining of the plurality of second part areas is based on a start part area, and

the start part area is based on the registration of the first medical image data and the second medical image data.

5. The computer-implemented method as claimed in claim 1 , wherein the determining synthetic image data, the determining a measure of similarity and the adjusting at least one parameter of the trainable function are carried out iteratively,

wherein in one iteration a start part area is based on the selected second part area of the previous iteration.

6. The computer-implemented method as claimed in claim 1 , wherein a start part area and each other second part area of the plurality of second part areas at least partly cover each other.

7. The computer-implemented method as claimed in claim 1 , further comprising:

receiving initial first medical image data, the initial first medical image data being based on the first medical imaging of the examination object;

determining initial synthetic image data by applying the trainable function to the initial first medical image data;

determining a first mapping function by registering the initial synthetic image data and the second medical image data; and

determining the first medical image data by applying the first mapping function to the initial first medical image data.

8. The computer-implemented method as claimed in claim 1 , wherein the determining synthetic image data, the determining a measure of similarity and the adjusting at least one parameter of the trainable function are carried out iteratively, the computer-implemented method further comprising:

determining a second mapping function by registering the synthetic image data and the second medical image data; and

determining the first medical image data for the following iteration by applying the second mapping function to the first medical image data.

9. The computer-implemented method as claimed in claim 1 , wherein

the trainable function comprises a Generative Adversarial Network,

the Generative Adversarial Network comprises a generator and a discriminator, and the generator advantageously comprises a Dense Unit Network (U net).

10. The computer-implemented method as claimed in claim 1 , wherein

the adjusting at least one parameter of the trainable function is based on an optimization of a target function,

the target function comprises a similarity function,

the measure of similarity depends on the similarity function, and

a weighting of the similarity function in the target function is variable.

11. The computer-implemented method as claimed in claim 10 , wherein

the determining synthetic image data, the determining a measure of similarity and the adjusting at least one parameter of the trainable function are carried out iteratively, and

the weighting of the similarity function increases with the iterations.

12. The computer-implemented method as claimed in claim 10 ,

wherein the weighting of the similarity function is adjusted based on a quality of the registration of the synthetic image data and the second medical image data,

wherein, the weighting of the similarity function increases based on the quality of the registration of the synthetic image data and the second medical image data.

13. A computer-implemented use of the trainable function provided in accordance with the computer-implemented method of claim 1 , comprising:

determining the synthetic image data by applying the trainable function to the first medical image data; and

providing the synthetic image data.

14. A non-transitory computer-readable memory medium storing instructions, when executed by a training system, are configured to cause the training system to perform the method of claim 1 .

15. A training system for providing a trainable function for determination of synthetic image data, comprising:

an interface; and

a computing device, the interface being configured to receive first medical image data, the first medical image data being based on a first medical imaging of an examination object,

the interface being further configured to receive second medical image data, the second medical image data being based on a second medical imaging of the examination object, the first and the second medical imaging differ by at least one of an imaging modality used or by an imaging protocol used,

the first and the second medical image data are registered with one another,

the computing device being configured to determine synthetic image data by applying the trainable function to the first medical image data,

the computing device being further configured to determine a measure of similarity with a similarity function by comparison of the synthetic image data and the second medical image data, the determination of the measure of similarity is based on an optimization method, and the optimization method comprises a maximization of the measure of similarity by at least one of the geometrical of photometric transformation of at least one of the synthetic image data or of the second medical image data, the optimization method including,

determining at least one first part area of the synthetic image data,

determining a plurality of second part areas of the second medical image data,

determining a plurality of part measures of similarity between the at least one first part area and the plurality of second part areas, and

determining the similarity function based on the plurality of part measures of similarity, wherein

the similarity function is based on a comparison of the first part area of the synthetic image data and a selected second part area of the second medical image data, and

the selected second part area corresponds to that area of the plurality of second part areas with the maximum part measure of similarity,

the computing device being further configured to adjust at least one parameter of the trainable function by optimization of the similarity function based on the measure of similarity, and

the interface being further configured to provide the trainable function.