Iterative image reconstruction

1. A system comprising:

a memory; and

one or more processors that, when executing instructions stored in the memory, are configured to perform operations comprising:

accessing a current structural estimate of a region of interest;

generating a first simulated X-ray measurement based on the current structural estimate of the region of interest;

receiving a first real X-ray measurement; and

generating an update to the current structural estimate of the region of interest as a function of the first simulated X-ray measurement and the first real X-ray measurement, the update being generated invariant on the current structural estimate.

2. The system of claim 1 , wherein the first real X-ray measurement is received from a cone-beam computed tomography (CBCT) system or computed tomography (CT) system, and wherein the operations further comprise:

computing the update as a derivative of a statistical objective function with respect to the first simulated X-ray measurement.

3. The system of claim 1 , wherein the operations further comprise:

linearly projecting the update to the current structural estimate into an image space to form a perturbation;

scaling the perturbation by a scalar for stability; and

subtracting the scaled perturbation from the current structural estimate to generate an updated structural estimate.

4. The system of claim 3 , wherein the operations further comprise:

applying at least one of regularization, momentum or denoising to the updated structural estimate.

5. The system of claim 1 , wherein the current structural estimate comprises an X-ray attenuation map that represents a three-dimensional (3D) model of the region of interest.

6. The system of claim 5 , wherein the X-ray attenuation map comprises a linear X-ray attenuation map.

7. The system of claim 1 , wherein generating the first simulated X-ray measurement based on the current structural estimate of the region of interest comprises applying the current structural estimate to a model that generates an expected output of a real X-ray measurement.

8. The system of claim 7 , wherein the current structural estimate comprises an X-ray attenuation map, and wherein the model generates the expected output of the real X-ray measurement, for a given measurement index i per number of detector elements in a cone-beam computed tomography (CBCT) system, in accordance with:

z i =b i *exp(−[ Ax] i )+ r i ,

where b is an incident intensity of flood field, A is a system projection matrix describing a combination of each image pixel at each detector element, x is the current X-ray attenuation map, exp is an exponential function, and r is background noise including scatter.

9. The system of claim 8 , wherein the operations further comprise re-estimating r representing the background noise including scatter based on the current structural estimate at each iteration of generating the update.

10. The system of claim 8 , wherein the model comprises a modeling function representing beam hardening from a polyenergetic source, the modeling function comprising a machine learning technique that is trained to establish a relationship between a training real X-ray measurement and a training known simulated X-ray measurement; a linear model; or a non-linear model that fits X-ray data to some nominal value comprising at least one of relative electron density, mass density, monoenergetic attenuation, proton stopping power, or bone mineral density.

11. The system of claim 1 , wherein the operations further comprise repeating the generating of a simulated X-ray measurement, receiving of a real X-ray measurement, and generating of an update to the current structural estimate for multiple sets of simulated and real X-ray measurements.

12. The system of claim 1 , wherein the operations further comprise:

accessing an updated structural estimate of the region of interest;

generating a second simulated X-ray measurement based on the updated structural estimate of the region of interest;

receiving a second real X-ray measurement; and

generating a further update to the updated structural estimate of the region of interest as a function of the second simulated X-ray measurement and the second real X-ray measurement.

13. The system of claim 1 , wherein the operations further comprise:

accessing an objective function comprising a negative log-likelihood (NLL) function;

computing a loss by applying the NLL function to a combination of the first simulated X-ray measurement and the first real X-ray measurement; and

in response to determining that the loss fails to satisfy a criterion, performing the update to the current structural estimate, the criterion comprising a difference between adjacent updates falling below a threshold, a number of iterations falling below a maximum iteration value, an elapsed time falling below a maximum time limit, or input requesting termination and display of a result.

14. The system of claim 1 , wherein the current structural estimate comprises an X-ray attenuation map, and wherein the update to the current structural estimate of the region of interest is computed in accordance with:

A T ( y /( b *exp(− Ax )+ r )−1),

where A is a system projection matrix describing a combination of each image pixel at each detector element, A T is a transpose of the system projection matrix, y is the first real X-ray measurement, b is an incident intensity of flood field, x is the X-ray attenuation map, exp is an exponential function, and r is background noise including scatter.

15. The system of claim 1 , wherein the operations further comprise:

receiving a plurality of real X-ray measurements;

partitioning the plurality of real X-ray measurements into measurement groups, a first measurement group of the measurement groups corresponding to a group of X-ray measurements used to update the current structural estimate, and a second measurement group of the measurement groups corresponding to a group of X-ray measurements that is skipped from updating the current structural estimate;

determining that the first real-X-ray measurement falls within the first measurement group; and

in response to determining that the first real-X-ray measurement falls within the first measurement group, performing the update to the current structural estimate.

16. The system of claim 1 , wherein the operations further comprise:

collecting a plurality of updates to the current structural estimate, each of the plurality of updates being associated with a respective iteration;

identifying a pattern of updates based on the collected plurality of updates; and

performing the update to the current structural estimate based on the identified pattern of updates.

17. A method comprising:

accessing a current structural estimate of a region of interest;

generating a first simulated X-ray measurement based on the current structural estimate of the region of interest;

receiving a first real X-ray measurement; and

generating an update to the current structural estimate of the region of interest as a function of the first simulated X-ray measurement and the first real X-ray measurement, the update being generated invariant on the current structural estimate.

18. The method of claim 17 , further comprising:

computing the update as a derivative of a statistical objective function with respect to the first simulated X-ray measurement.

19. The method of claim 17 , further comprising:

linearly projecting the update to the current structural estimate into an image space to form a perturbation;

scaling the perturbation by a scalar for stability; and

subtracting the scaled perturbation from the current structural estimate to generate an updated structural estimate.

20. The method of claim 19 , further comprising:

applying at least one of regularization, momentum or denoising to the updated structural estimate.

21. The method of claim 17 , wherein the current structural estimate comprises an X-ray attenuation map that represents a three-dimensional (3D) model of the region of interest.

22. The method of claim 21 , wherein the X-ray attenuation map comprises a linear X-ray attenuation map.

23. The method of claim 17 , wherein generating the first simulated X-ray measurement based on the current structural estimate of the region of interest comprises applying the current structural estimate to a model that generates an expected output of a real X-ray measurement.

24. A non-transitory computer-readable medium comprising non-transitory computer-readable instructions that, when executed by one or more processors, configure the one or more processors to perform operations comprising:

accessing a current structural estimate of a region of interest;

generating a first simulated X-ray measurement based on the current structural estimate of the region of interest;

receiving a first real X-ray measurement; and

generating an update to the current structural estimate of the region of interest as a function of the first simulated X-ray measurement and the first real X-ray measurement, the update being generated invariant on the current structural estimate.

25. The non-transitory computer-readable medium of claim 24 , wherein the operations further comprise:

computing the update as a derivative of a statistical objective function with respect to the first simulated X-ray measurement.

26. The non-transitory computer-readable medium of claim 24 , wherein the operations further comprise:

linearly projecting the update to the current structural estimate into an image space to form a perturbation;

scaling the perturbation by a scalar for stability; and

subtracting the scaled perturbation from the current structural estimate to generate an updated structural estimate.

27. The non-transitory computer-readable medium of claim 26 , wherein the operations further comprise:

applying at least one of regularization, momentum or denoising to the updated structural estimate.

28. The non-transitory computer-readable medium of claim 24 , wherein the current structural estimate comprises an X-ray attenuation map that represents a three-dimensional (3D) model of the region of interest.

29. The non-transitory computer-readable medium of claim 28 , wherein the X-ray attenuation map comprises a linear X-ray attenuation map.

30. The non-transitory computer-readable medium of claim 24 , wherein generating the first simulated X-ray measurement based on the current structural estimate of the region of interest comprises applying the current structural estimate to a model that generates an expected output of a real X-ray measurement.