A method for calibrating X-ray projection geometry in X-ray cone-beam computed tomography

Abstract

The present invention relates to a method for X-ray projection geometry calibration in X-ray cone beam computed tomography, comprising at least one step (S1) of acquiring a two-dimensional X-ray image (1) or sinogram (2) of at least a part of an object (3) generated by relatively rotating a detector and an X-ray source projecting X-rays towards the detector around the object (3), characterized in that the method further comprises at least one step (S4) of detecting at least one feature (3 a) of the object (3) in the two-dimensional X-ray image (1) or sinogram (2) by using a trained artificial intelligence algorithm, and at least one step (S5) of generating calibration information defining the geometry of the X-ray projection based on the detection.

Description

[Technical field]

[0001] The present invention relates to X-ray volume tomography (VT), and in particular to X-ray dental volume tomography (DVT). The present invention more particularly relates to a method for calibrating X-ray projection geometry in X-ray volume tomography. The present invention more particularly relates to motion artifact reduction in X-ray volume tomography. The present invention more particularly relates to X-ray cone beam computed tomography (CBCT). [Background technology]

[0002] In systems for X-ray cone-beam computed tomography, a sensor and a detector rotate relatively around a patient, in particular around the patient's head, to generate a number of projection images, i.e. two-dimensional (2D) X-ray images. In order to calculate a three-dimensional (3D) tomographic image from these 2D X-ray images, the projection geometry of each 2D X-ray image must be known. Calibration of the projection geometry of an X-ray CBCT system is usually performed in the factory or after the X-ray CBCT system is installed in the clinic. For this purpose, 2D X-ray images are generated using a calibration body. The calibration body has a predefined shape and one or more features that are easy to detect and identify in the 2D X-ray image. Furthermore, the relative positions of the features are generally known. By applying so-called consistency conditions to the 2D X-ray image, the projection geometry of the 2D X-ray image can be derived. In the literature “Auto-calibration by locally consistent contours for dental CBCT, S. Maur et al 2018 Phys. Med. Boil. 63 215018 , pp. 1111-1123, 2002. The projection geometry forms the basis of the calibration information of the X-ray CBCT system. Mechanical changes such as movements and / or deformations in the X-ray CBCT system relative to the irradiated patient can invalidate the calibration information. For example, a relative movement of the patient during the acquisition of the 2D X-ray images results in an inaccurate calibration information of the X-ray CBCT system. For accurate reconstruction of 3D tomographic images, the projection geometry is relative to the patient, not to a fixed point in the room, such as a calibration body. If the patient moves during the acquisition, the assumed X-ray projection geometry will not match the acquired 2D X-ray images. This leads to data inconsistencies in the reconstruction of the 3D tomographic images. This therefore further results in motion artifacts in the 3D tomographic images, such as blurred structures and / or double edges. Therefore, in the prior art, automatic calibration methods have been envisaged to create calibration information for 2D X-ray images. However, these are still the subject of research and development. In such automatic calibration methods, any part of the patient may be used as a feature for the calibration, for example, anatomical parts of the patient may be used.However, accurate detection of these features in 2D x-ray images is difficult, since their shapes, positions, and densities are generally unknown. Feature detection is complicated by many overlapping anatomical parts, and low contrast at low x-ray doses as well as high image noise. Thus, the information content of individual projection images is usually not sufficient for accurate feature detection. To increase accuracy, prior knowledge can be incorporated into the feature detection. For example, the prior knowledge can relate to the shape, position, and density of the anatomical parts. The prior knowledge can further relate to assumptions about the movement of the patient or the x-ray CBCT system. However, complete modeling of such prior knowledge and its integration into the feature detection algorithm is not trivial, and high accuracy and completeness are difficult to achieve. This makes image processing for detecting features in 2D x-ray images very slow and prone to errors, and thus the accuracy of the calibration information and the speed of the automatic calibration process are adversely affected. Summary of the Invention

[0003] It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and to provide a method for projection geometry calibration in X-ray cone-beam computed tomography.

[0004] This object is achieved by an X-ray projection geometry calibration method according to claim 1 and an X-ray cone-beam computed tomography system according to claim 13. The subject matter of the dependent claims relates to further developments.

[0005] The present invention provides a method for X-ray projection geometry calibration in X-ray cone beam computed tomography, comprising at least one step of acquiring two-dimensional X-ray images or sinograms of at least a portion of an object, generated by relatively rotating a detector and an X-ray source projecting X-rays towards the detector around the object, at least one step of detecting features of the object in at least one of the two-dimensional X-ray images or sinograms by using a trained artificial intelligence algorithm, and at least one step of generating calibration information defining the geometry of the X-ray projections based on the detection.

[0006] The main advantage of the present invention is that a trained artificial intelligence algorithm is applied to feature detection in 2D X-ray images or sinograms to calibrate the X-ray projection geometry. The artificial intelligence algorithm learns the 2D / 3D positions, shapes, and even densities of anatomical parts, and the implicit knowledge of the data consistency conditions, as well as the typical motion effects of patients and X-ray CBCT systems. This makes the artificial intelligence algorithm more accurate and much faster than classical image processing. This makes it possible to more effectively reduce motion artifacts in 3D tomographic images, and obtain sharp 3D tomographic images.

[0007] According to the invention, features of objects in 2D X-ray images or sinograms can be detected via 2D or 3D masks generated by a trained artificial intelligence algorithm. The 2D or 3D masks represent the location / shape of the features, in particular the contour or silhouette. The 2D and 3D masks can be binary masks. Alternatively, probability or confidence masks may be used.

[0008] According to the invention, the created calibration information can be X-ray unit specific or patient specific calibration information. X-ray unit specific calibration information can be created by irradiating a calibration object such as a dummy. Patient specific calibration information can be created by irradiating a patient. The patient feature to be detected can correspond to an anatomical part or a prosthetic part. The anatomical part can be a tooth or a bone. And the prosthetic part can be an implant, a crown, a filling, etc.

[0009] According to the invention, X-ray unit-specific calibration information for an X-ray CBCT system may be created at the factory when the system is manufactured and then temporarily or permanently stored in the X-ray CBCT system. Alternatively, X-ray unit-specific calibration information may be created during installation and optionally at predetermined intervals after installation and also stored in the X-ray CBCT system. X-ray unit-specific calibration information may also be created and stored when an artifact is observed in a computed tomography. Patient-specific calibration information for an X-ray CBCT system may be created separately for each patient and then stored for use when reconstructing a volume tomographic image for each patient.

[0010] According to the invention, various patient-specific calibration information for different patients can be created, stored and compared with the X-ray unit-specific calibration information to analyze discrepancies and draw conclusions regarding the accuracy of the X-ray unit-specific calibration or the patient's motion behavior. Inaccurate X-ray unit-specific calibration information can be automatically adjusted based on the analysis. Alternatively, to initiate the creation of new calibration information, the user can be informed about defects in the X-ray unit-specific calibration information by displaying the defect information on the control display or on the display of the reconstructed 3D tomographic image. Alternatively, the user can be informed via a speaker. Alternatively, the user can be informed about the patient's motion behavior as well, preferably when such motion is excessive and motion artifacts cannot be reduced satisfactorily.

[0011] In accordance with the present invention, previously stored x-ray unit specific calibration information may be updated with recently created x-ray unit specific calibration information. Similarly, previously stored patient specific calibration information may be updated with recently created patient specific calibration information corresponding to the same patient.

[0012] According to the invention, a three-dimensional tomographic image of at least a part of a patient's body may be reconstructed using X-ray unit specific calibration information or alternatively corresponding patient specific calibration information.

[0013] According to the present invention, the two-dimensional X-ray image or sinogram can be optionally pre-processed by image processing based on trained artificial intelligence algorithms or classical algorithms before the detection step. The image processing algorithms can include at least a filtering process, a contrast enhancement process, an edge enhancement process, and a noise suppression process.

[0014] According to the invention, one or more or even all of the features to be detected may be selected automatically or, alternatively, may be selected manually by a user on the display.

[0015] According to the invention, the artificial intelligence algorithm for detecting features is preferably trained by using data pairs. Each data pair includes a 2D X-ray image and an associated 2D mask representing the 2D position and / or shape of the feature in the associated 2D X-ray image. Alternatively, each data pair includes a sinogram and an associated 3D mask representing the 3D position and / or shape of the feature in the sinogram. These data pairs may be generated by robotic movement of an actual calibration body in a laboratory. Alternatively, these data pairs may be acquired from different patients. Alternatively, the data pairs may be generated virtually by simulation.

[0016] The invention also provides an X-ray volume tomography system, preferably an X-ray cone-beam computed tomography system. The X-ray cone-beam computed tomography system comprises an X-ray unit with generating means for generating a two-dimensional X-ray image or sinogram of at least a part of an object by relatively rotating around the object an X-ray source for projecting X-rays towards the detector and a detector. The object may be a calibration body such as a dummy or a patient, preferably a patient's head, more preferably a patient's jaw. The X-ray volume tomography system further comprises a tomographic reconstruction unit with image processing means adapted to perform the steps of the X-ray projection geometry calibration method of the invention.

[0017] The invention also provides a computer readable program comprising code for causing a computer-based X-ray volume tomography system to carry out the method steps of the invention. The computer readable program may be provided together with the X-ray volume tomography system or on a computer readable storage device.

[0018] In the following description, the invention is explained in more detail using exemplary embodiments and by referring to the following drawings: [Brief description of the drawings]

[0019] [Figure 1] 3 is a flow chart showing the steps of a method for X-ray projection geometry calibration in X-ray cone beam computed tomography according to the present invention; [Diagram 2] 2 is a two-dimensional X-ray image produced via an X-ray unit according to the present invention. [Diagram 3] 3 is a cross-sectional view of a sinogram taken along line AA in FIG. 2 of a two-dimensional X-ray image produced via an X-ray unit according to the present invention;

[0020] The reference numbers shown in the drawings refer to the elements listed below and are referenced in the following description of the exemplary embodiments. 1.2DX ray image 2. Sinogram 3. Object (e.g., patient or calibration object) 3a. Features (e.g. anatomical / prosthetic parts) 3b.Position / shape DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] FIG. 1 shows steps (S1 to S7) of an X-ray projection geometry calibration method for X-ray cone beam computed tomography according to one embodiment of the present invention.

[0022] In a first step (S1), multiple two-dimensional X-ray images (1) or sinograms (2) of at least a portion of an object (3) are acquired from an X-ray cone-beam computed tomography system of the present invention. The object (3) can be a patient or a calibration object such as a dummy. The three-dimensional X-ray images (1) and sinograms (2) are generated by the X-ray cone-beam computed tomography system by relatively rotating a detector and an X-ray source that projects X-rays towards the detector around the object (3).

[0023] In an optional second step (S2), the 2D X-ray image (1) or sinogram (2) is pre-processed. The pre-processing step (S2) comprises at least one of a filtering process, a contrast enhancement process, an edge enhancement process and a noise suppression process.

[0024] In another optional third step (S3), one or more features (3a) of the object (3) are selected in the originally acquired or pre-processed 2D X-ray image (1) or sinogram (2). The selection can be done manually on the display. Alternatively, the selection is done automatically.

[0025] In a fourth step (S4), at least one feature (3a) of the object (3), optionally a selected feature (3a), is detected in the 2D X-ray image (1) or sinogram (2) or, optionally, in the preprocessed 2D X-ray image (1) or preprocessed sinogram (2) by using a trained artificial intelligence algorithm. Optionally, additional information may be input to the trained artificial intelligence algorithm for detecting the feature (3a) in the fourth step (S4). The additional information may include an initial estimate for the 2D / 3D position, shape and relative movement of the feature (3a). The additional information may include information for improving the detection of the at least one selected feature (3a) in step (S3). The features (3a) may be detected simultaneously or one by one. After the selection step (S3), a further preprocessing step can optionally be performed. Thereby, for example, the preprocessing can be adapted to the spatial or image characteristics of the selected feature (3a) to optimize the feature detection. The features (3a) to be detected may correspond to anatomical or prosthetic parts of the patient, in particular of the patient's head, more particularly of the patient's jaw. The anatomical parts may be teeth or bones, etc. The prosthetic parts may be implants, etc. As shown in FIG. 2, the trained artificial intelligence algorithm generates a 2D mask representing the 2D position and / or 2D shape (3b) of the features (3a) of the object (3) in the 2D X-ray image (1), i.e. the teeth in the patient's jaw. FIG. 3 is a cross-sectional view of the sinogram (2) along line AA in FIG. 2, including the 2D X-ray image (1) of FIG. 2 at the projection index "100". The projection index and coordinates may vary depending on the tomography details. The projection index shows a number of 2D X-ray images (1) in the sinogram (2). As shown in FIG. 3, the trained artificial intelligence algorithm may alternatively generate a 3D mask representing the 3D position and / or 3D shape (3b) of the features (3a) of the object (3) in the sinogram (2), i.e. the teeth in the patient's jaw. In the following description, reference numbers with a prime sign indicate elements that are used to train an artificial intelligence algorithm. The artificial intelligence algorithm is trained using data pairs.Each data pair includes a 2D X-ray image (1') and an associated 2D mask representing the position and / or shape (3b') of a feature (3a') in the associated 2D X-ray image (1). Alternatively, each data pair includes a sinogram (2') and an associated 3D mask representing the position and / or shape (3b') of a feature (3a') in the sinogram (2).

[0026] In a fifth step (S5), based on the detection, calibration information is generated. The calibration information defines the geometry of the X-ray projection. When the irradiated object (3) is a calibration body, the generated calibration information corresponds to X-ray unit-specific calibration information. When the irradiated object (3) is a patient, the generated calibration information corresponds to patient-specific calibration information.

[0027] In an optional first step (S6a) of the sixth step, recently created X-ray unit-specific calibration information and patient-specific calibration information are temporarily or permanently stored in the X-ray cone beam computed tomography system. Different patient-specific calibration information for different patients may be created and stored. In an optional subsequent step (S6b) of the sixth step, previously stored X-ray unit-specific calibration information and previously stored patient-specific calibration information are updated with recently created X-ray unit-specific calibration information and recently created patient-specific calibration information of the same patient, respectively.

[0028] In a seventh step (S7), a three-dimensional tomographic image is reconstructed based on the two-dimensional X-ray image (1) or sinogram (2) of at least a part of the patient's body and on the X-ray unit specific calibration information. Alternatively, the three-dimensional tomographic image may be reconstructed based on the two-dimensional X-ray image (1) or sinogram (2) of at least a part of the patient's body and on the corresponding patient specific calibration information.

[0029] The inventive X-ray cone-beam computed tomography system comprises an X-ray unit and a tomographic reconstruction unit. The X-ray unit comprises acquisition means adapted to acquire a two-dimensional X-ray image (1) or sinogram (2) of at least a part of an object (3) by relatively rotating an X-ray source and a detector around the object (3). The tomographic reconstruction unit comprises image processing means adapted to perform the steps of an X-ray projection geometry calibration method.

[0030] The method may be provided in the form of a computer readable program having code for causing a computer-based X-ray cone beam computed tomography system to perform the above-mentioned method steps (S1-S7). The computer readable program may be stored in a computer readable storage device of the computer-based X-ray cone beam computed tomography system. The following is a summary of the claims as originally filed. [1] A method for X-ray projection geometry calibration in X-ray cone beam computed tomography, comprising: A method comprising at least one step (S1) of acquiring a two-dimensional X-ray image (1) or sinogram (2) of at least a part of an object (3), generated by relatively rotating around the object (3) a detector and an X-ray source projecting X-rays towards said detector, at least one step (S4) of detecting in said 2D X-ray image (1) or said sinogram (2) at least one feature (3a) of said object (3) by using a trained artificial intelligence algorithm; - at least one step (S5) of generating calibration information defining the geometry of said X-ray projections based on said detection; The method of claim 1, further comprising: [2] The method according to [1], characterized in that in the detection step (S4), the trained artificial intelligence algorithm generates a 2D mask representing the 2D position and / or 2D shape (3b) of the feature (3a) of the object (3) in the 2D X-ray image (1) or a 3D mask representing the 3D position and / or 3D shape of the feature (3a) of the object (3) in the sinogram (2). [3] The method according to [1] or [2], characterized in that in at least one acquisition step (S1), the object (3) is a calibration object and the created calibration information corresponds to calibration information specific to an X-ray unit. [4] A method according to any one of [1] to [3], characterized in that in at least one acquisition step (S1), the object (3) is a patient, the corresponding feature (3a) comprises an anatomical or prosthetic part of the patient's body or the patient's head, and the created calibration information corresponds to patient-specific calibration information. [5] The method according to [4], characterized in that the anatomical part comprises at least a part of a tooth or bone and / or the prosthetic part comprises at least a part of an implant. [6] The method, - storing (S6a) the recently created X-ray unit specific calibration information and / or the recently created patient specific calibration information, or updating (S6b) the previously stored X-ray unit specific calibration information and / or the previously stored patient specific calibration information with the recently created X-ray unit specific calibration information and / or the recently created patient specific calibration information, respectively; The method according to any one of [3] to [5], further comprising: [7] A method according to any one of [3] to [6], further comprising a step (S7) of reconstructing a three-dimensional tomographic image based on the two-dimensional X-ray image (1) or sinogram (2) of at least a part of the patient's body and calibration information specific to the X-ray unit. [8] The method according to any one of [3] to [6], further comprising a step (S7) of reconstructing a three-dimensional tomographic image based on the two-dimensional X-ray image (1) or the sinogram (2) of at least a part of the patient's body and the corresponding patient-specific calibration information. [9] The method according to any one of [1] to [8], further comprising a step (S2) of preprocessing the two-dimensional X-ray image (1) or the sinogram (2) prior to the detection step (S4).

[10] The method according to any one of [1] to [9], further comprising a step (S3) of selecting one or more features (3a) to be detected in the detection step (S4).

[11] The method according to [9], characterized in that the pre-processing step (S2) comprises at least one of a filtering process, a contrast enhancement process, an edge enhancement process and a noise suppression process.

[12] A method according to any one of [1] to

[11] , further comprising a step of training the artificial intelligence algorithm by using data pairs, characterized in that each data pair comprises a 2D X-ray image (1') and an associated 2D mask representing the 2D position and / or 2D shape (3b') of a feature (3a') in the associated 2D X-ray image (1), or each data pair comprises a sinogram (2) and an associated 3D mask representing the 3D position and / or 3D shape (3b') of a feature (3a') in the sinogram (2).

[13] An X-ray cone beam computed tomography system, comprising: 1. An X-ray cone beam computed tomography system comprising an X-ray unit with acquisition means adapted to acquire a two-dimensional X-ray image (1) or sinogram (2) of at least a part of an object (3) by relatively rotating an X-ray source and a detector around said object (3), The tomographic reconstruction unit further comprises an image processing means, the tomographic reconstruction unit comprising: detecting at least one feature (3a) of said object (3) in said 2D X-ray image (1) or said sinogram (2) by using a trained artificial intelligence algorithm; generating calibration information defining a geometry of the X-ray projections based on the detection; and 1. An X-ray cone beam computed tomography system, comprising:

[14] A computer-readable program comprising code for causing a computer-based X-ray volume tomography system to execute the steps of the method according to any one of [1] to

[12] .

[15] A computer-readable storage device storing the computer-readable program described in

[14] .

Claims

1. 1. A method for X-ray projection geometry calibration in X-ray cone beam computed tomography, comprising:

1. A method comprising at least one step (S1) of acquiring a sinogram (2) of at least a part of an object (3), generated by relatively rotating a detector and an X-ray source projecting X-rays towards said detector around said object (3), at least one step (S4) of detecting in said sinogram (2) at least one feature (3a) of said object (3) by using a trained artificial intelligence algorithm, wherein in at least one acquisition step (S1), said object (3) is a patient and said corresponding feature (3a) comprises an anatomical or prosthetic part of said patient's body or of said patient's head, at least one step (S5) of generating calibration information defining a geometry of said X-ray projections based on said detection, wherein said generated calibration information corresponds to patient-specific calibration information; - reconstructing (S7) a three-dimensional tomographic image based on the sinogram (2) of at least a portion of the patient's body and on the corresponding patient-specific calibration information; The method of claim 1, further comprising:

2. 2. The method according to claim 1, characterized in that in the detection step (S4), the trained artificial intelligence algorithm generates a 3D mask representing the 3D position and / or the 3D shape of the features (3a) of the object (3) in the sinogram (2).

3. 3. The method according to claim 1 or 2, characterized in that the anatomical part comprises at least a part of a tooth or a bone and / or the prosthetic part comprises at least a part of an implant.

4. The method further comprising: storing recently created patient-specific calibration information (S6a); or updating (S6b) each previously stored patient-specific calibration information with the recently created patient-specific calibration information; The method according to any one of claims 1 to 3, further comprising:

5. The method according to any one of claims 1 to 4, characterized in that it further comprises a step (S2) of pre-processing the sinogram (2) before the detection step (S4).

6. Method according to any one of the preceding claims, characterized in that it further comprises a step (S3) of selecting one or more features (3a) to be detected in said detection step (S4).

7. The method according to claim 5, characterized in that the pre-processing step (S2) comprises at least one of the following processes: a filtering process, a contrast enhancement process, an edge enhancement process and a noise suppression process.

8. The method according to any one of claims 1 to 7, further comprising the step of training the artificial intelligence algorithm by using data pairs, each data pair comprising a sinogram (2) and an associated 3D mask representing the 3D position and / or the 3D shape (3b') of a feature (3a') in the sinogram (2).

9. 1. An x-ray cone beam computed tomography system, comprising:

1. An X-ray cone-beam computed tomography system comprising an X-ray unit with acquisition means adapted to acquire a sinogram (2) of at least a part of an object (3) by relatively rotating an X-ray source and a detector around said object (3), The tomographic reconstruction unit further comprises an image processing means, the tomographic reconstruction unit comprising: Executing the steps of the method according to any one of claims 1 to 8 by using a trained artificial intelligence algorithm.

1. An X-ray cone beam computed tomography system, comprising:

10. A computer readable program comprising code for causing a computer-based X-ray volume tomography system to carry out the steps of the method according to any one of claims 1 to 8.

11. A computer readable storage device storing the computer readable program of claim 10.

Images