Learning-based correction of grid artifacts in X-ray imaging
A method for training a function of an X-ray system that has a positioning mechanism such as a C-arm, a detector, and, in a beam path in front of the detector, an anti-scatter grid. Positioning of the detector at a large number of different positions occurs. The positioning mechanism is deflected and/or distorted. Recording of at least one X-ray photograph in each of the positions then takes place, and the method further includes machine learning of artifacts generated by the anti-scatter grid from all X-ray photographs for the function.
1. A method for training a function of an X-ray system, the X-ray system including a positioning mechanism, a detector, and, in a beam path in front of the detector, an anti-scatter grid, the method comprising:
positioning the detector at a number of different positions, wherein the positioning mechanism is deflected, distorted, or deflected and distorted;
recording in each case at least one X-ray photograph in each of the number of different positions; and
machine learning artifacts generated by the anti-scatter grid from all of the X-ray photographs for the function in a first learning step.
2. The method of claim 1 , wherein on each positioning of the detector, respective system geometry data of the X-ray system is supplied, the system geometry data serving as input variables for machine learning.
3. The method of claim 1 , wherein each of the number of different positions of the detector for machine learning is approached multiple times by the positioning mechanism.
4. The method of claim 3 , wherein each of the number of different positions is approached from different directions.
5. The method of claim 1 , wherein the number of different positions of the detector is arranged uniformly distributed over an entire system-based movement space of the X-ray system.
6. The method of claim 1 , wherein the machine learning, in addition to the first learning step, comprises a second learning step, the second learning step being identical to the first learning step, and
wherein the anti-scatter grid is removed from the detector and removed from the beam path in the second learning step.
7. The method of claim 1 , wherein on recording the X-ray photographs, a phantom that does not generate any scatter radiation is placed in the beam path.
8. The method of claim 1 , further comprising obtaining virtual training data by way of simulation,
wherein, in addition to the X-ray photographs, the obtained virtual training data is used for machine learning.
9. The method of claim 1 , wherein an object X-ray photograph is obtained from an object, and artifacts of the anti-scatter grid are reduced or eliminated in the object X-ray photograph by the trained function.
10. The method of claim 1 , wherein the positioning mechanism includes a C-arm.
11. A method for operating an X-ray system, the X-ray system including a positioning mechanism, a detector, and in a beam path in front of the detector, an anti-scatter grid, the method comprising:
positioning the detector at a position, wherein the positioning mechanism is deflected, distorted, or deflected and distorted;
recording an X-ray photograph in the position; and
correcting artifacts resulting, which are caused by the anti-scatter grid, in the X-ray photograph with deflection, distortion, or deflection and distortion of the positioning mechanism by artifacts learned according to a method for training a function of the X-ray system, the method for training the function of the X-ray system comprising:
positioning the detector at a number of different positions, wherein the positioning mechanism is deflected, distorted, or deflected and distorted;
recording in each case at least one X-ray photograph in each of the number of different positions; and
machine learning artifacts generated by the anti-scatter grid from all of the X-ray photographs for the function in a first learning step.
12. The method of claim 11 , wherein the positioning mechanism comprises a C-arm.
13. A computer-implemented method for generating object image data, the computer-implemented method comprising:
receiving an object X-ray photograph;
generating a corrected object-X-ray photograph, the generating comprising applying a trained function to the object-X-ray photograph, such that artifacts of the anti-scatter grid are reduced or eliminated in the object X-ray photograph; and
supplying the corrected object X-ray photograph as the object image data.
14. An X-ray system comprising:
a positioning mechanism;
a recording device including a detector and an anti-scatter grid located in a beam path in front of the detector, wherein the detector and the anti-scatter grid are attached to the positioning mechanism, wherein with the positioning mechanism, the detector is positionable at a large number of different positions in which the positioning mechanism is deflected, distorted, or deflected and distorted as a function of position, and wherein with the recording device, in each case at least one X-ray photograph is generatable in each of the number of different positions; and
a processor configured for machine learning of artifacts generated by the anti-scatter grid from all the X-ray photographs in a first learning step.
15. The X-ray system of claim 14 , wherein the X-ray system is configured as a C-arm X-ray system or a robotic arm-based X-ray system.