CN111369635A - System and method for predicting truncated image, method for preparing data, and medium thereof - Google Patents

Abstract

The present invention relates to a system and method for predicting a truncated image, a method for preparing data, and a medium thereof. The method for preparing data includes a virtual simulation step for performing virtual simulation on image data, so as to simultaneously obtain a virtual distorted image with data truncation and a virtual refined standard image without data truncation. The method for predicting a truncated image predicts a truncated image based on a trained learning network, and the trained learning network is obtained by performing data training based on a data set consisting of a virtual distorted image and a virtual refined standard image obtained by the above-mentioned method of preparing data. . Also provided are a system corresponding to the above method and a recording medium in which the recorded instructions can implement the above method.

Description

System and method for predicting truncated image, method for preparing data and medium thereof

technical field

The present invention relates to the field of medical imaging, and in particular to techniques for preparing data for predicting truncated images in computed tomography (CT) imaging and predicting truncated images based on the data.

Background technique

During a CT scan, a detector is used to collect X-ray data after passing through the patient's body, and then the collected X-ray data is processed to obtain projection data. These projection data can be used to reconstruct slice images. Complete projection data reconstructs accurate slice images for diagnosis.

However, if the patient is large or in a special position, some parts of the patient's body will be out of the scanning field, and the detector will not be able to collect the complete projection data, which is called data truncation. This data truncation introduces truncation artifacts and ultimately results in distorted reconstructed images. Distorted images are definitely not ideal in radiation therapy, because doctors must know the lines of the body skin and the number of CTs along the beam when making a diagnosis, so that they can accurately judge the radiation dose that should be applied to the patient, but Distorted images do not accurately reflect the above information. In this way, how to recover the image outside the scanning domain (we call it the truncated image) is a problem that must be solved.

There are several traditional methods for dealing with the above truncation problem. They use some mathematical models to predict the truncated projection data, such as using the water model to predict the truncated part, but the quality of the truncated image recovered by these traditional methods will vary. The actual situation changes, and the performance is not ideal. In addition, different users often involve different sets of patients, and the image data they need is also focused, but the traditional method never involves classifying the image data.

In recent years, a new technique has emerged in which truncated images are predicted by means of artificial intelligence (AI). Predicting truncated images through AI undoubtedly has huge advantages that traditional technologies cannot match. However, the performance of AI depends on its input data. Hospitals or research institutions and departments continue to accumulate raw image data every day, but we obviously cannot simply input these accumulated raw image data into the AI network for learning, because the performance of AI depends on the quality of the input data, not the quality of the input data. quantity. On the other hand, to predict truncated images through AI, the input data set corresponding to the distorted image and the precise standard data set corresponding to the precise standard image without data truncation must be input into the AI network at the same time. However, such an image is not easy. Get, or, the data type is also relatively single.

SUMMARY OF THE INVENTION

One objective of the present invention is to overcome the above-mentioned and/or other problems in the prior art, which can obtain sufficient and reliable data for algorithm verification, thereby greatly helping to improve the accuracy of prediction of truncated images.

According to a first aspect of the present invention, there is provided a method for preparing data for predictive truncated images, comprising a virtual simulation step for virtual simulation of image data to obtain simultaneously a virtual distorted image with data truncated and without data Truncated virtual fine standard image.

Preferably, before the virtual simulation step, the method further includes an adaptive classification step, the adaptive classification step is used for adaptively classifying the collected image data according to predefined features, and the virtual simulation step for virtual simulation of the classified image data.

The predefined characteristics may include image data types. Further, the predefined features may further include possibilities corresponding to the image data types.

The image data type may be a patient's anatomy.

Preferably, the virtual simulation step further comprises: receiving an original image without data truncation; virtual shifting (offset) part of the original image corresponding to the target object to partially move out of the scanning domain, thereby obtaining a virtual refined standard image; Perform a simulated scan on the virtual fine standard image and perform virtual data acquisition to generate virtual truncated data; and perform image reconstruction processing on the virtual truncated data to obtain a virtual distorted image.

Preferably, the virtual simulation step further comprises: receiving an original image without data truncation, the original image being used as a virtual refined standard image; keeping the original image in the scanning domain, and performing orthographic projection on the original image processing to obtain a sinogram of the original image; excising the left and right channels of the sinogram and padding with padding values to generate virtual truncated data; and performing image reconstruction processing on the virtual truncated data to obtain Virtually distorted image.

More preferably, the above-mentioned fill value is an edge fill value.

According to a second aspect of the present invention, there is provided a method for predicting a truncated image, comprising the steps of: predicting a truncated image based on a trained learning network obtained by using the above-described method of preparing data The data set composed of virtual distorted images and virtual refined standard images is obtained by data training.

According to a third aspect of the present invention, there is provided a system for predicting a truncated image, comprising: a virtual simulation device for performing virtual simulation on image data to simultaneously obtain a virtual distorted image with data truncated and a virtual distorted image without data truncated A refined standard image; and a prediction device for predicting a truncated image based on a trained learning network obtained by data training based on a data set formed by the virtual distorted image and the virtual refined standard image.

Preferably, the system further includes an adaptive classifier for adaptively classifying the collected image data according to predefined features, and the virtual simulation device is used for virtual simulation of the classified image data.

Preferably, the virtual simulation device is further configured to: receive an original image without data truncation; virtual shift (offset) part of the original image corresponding to the target object to partially move out of the scanning domain, thereby obtaining a virtual precision standard. image; perform simulated scanning on the virtual fine standard image and perform virtual data acquisition to generate virtual truncated data; and perform image reconstruction processing on the virtual truncated data to obtain a virtual distorted image.

Preferably, the virtual simulation device is further configured to: receive an original image without data truncation, the original image is used as a virtual refined standard image; keep the original image in the scanning domain, and perform a Orthographic processing to obtain a sinogram of the original image; excising the left channel and right channel of the sinogram, and filling with padding values to generate virtual truncated data; and performing image reconstruction processing on the virtual truncated data, to get a virtual distorted image.

According to a fourth aspect of the present invention, there is also provided a computer-readable storage medium, the instructions recorded thereon can implement the above method and system.

According to the method for preparing data for predicting a truncated image, the real data truncation can be simulated intelligently to obtain a distorted image and a fine standard image required for the prediction of the truncated image; The data is intelligently classified into different categories, which is more beneficial to the above-mentioned distorted images and refined standard images.

The method and system for predicting a truncated image according to the present invention predicts a truncated image based on a trained learning network, which is obtained by data training based on the data set consisting of the above-mentioned virtual distorted image and virtual refined standard image. , so that the prediction result can be obtained quickly and more accurately.

Other features and aspects will become apparent from the following detailed description, drawings, and claims.

Description of drawings

The present invention may be better understood by describing exemplary embodiments of the present invention in conjunction with the accompanying drawings, in which:

1 is a flowchart of a method for preparing data for prediction of truncated images according to an exemplary embodiment of the present invention;

Fig. 2 is the flow chart of the first embodiment of the virtual simulation step in the method shown in Fig. 1;

Fig. 3 is the schematic diagram of the first embodiment of the virtual simulation step in the method shown in Fig. 1;

Fig. 4 is the flow chart of the second embodiment of the virtual simulation step in the method shown in Fig. 1;

Fig. 5 is the schematic diagram of the second embodiment of the virtual simulation step in the method shown in Fig. 1;

6 is a flowchart of an alternative embodiment of a method for preparing data for prediction of truncated images according to an exemplary embodiment of the present invention;

Fig. 7 shows an example of data classification in the adaptive classification step in the method shown in Fig. 6;

8 is a flowchart of a method for predicting a truncated image according to an exemplary embodiment of the present invention;

FIG. 9 is a schematic block diagram of a system for predicting a truncated image according to an exemplary embodiment of the present invention; and

FIG. 10 shows an example of a system for predicting a truncated image according to an exemplary embodiment of the present invention.

specific embodiment

The specific embodiments of the present invention will be described below. It should be noted that, in the specific description of these embodiments, for the sake of brevity and conciseness, this specification may not describe all the features of the actual embodiments in detail. It should be understood that, in the actual implementation process of any embodiment, just as in the process of any engineering project or design project, in order to achieve the developer's specific goals, in order to meet the system-related or business-related constraints, Often a variety of specific decisions are made, which also vary from one implementation to another. Furthermore, it will also be appreciated that while such development efforts may be complex and tedious, for those of ordinary skill in the art to which this disclosure pertains, the techniques disclosed in this disclosure will Some changes in design, manufacture or production based on the content are only conventional technical means, and it should not be understood that the content of the present disclosure is insufficient.

Unless otherwise defined, technical or scientific terms used in the claims and the specification shall have the ordinary meaning as understood by those with ordinary skill in the technical field to which this invention belongs. The terms "first", "second" and similar terms used in the description of the patent application and the claims of the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. "A" or "an" and the like do not denote a quantitative limitation, but rather denote the presence of at least one. Words like "including" or "comprising" mean that the elements or items appearing before "including" or "including" cover the elements or items listed after "including" or "including" and their equivalents, and do not exclude other components or objects. Words like "connected" or "connected" are not limited to physical or mechanical connections, nor are they limited to direct or indirect connections.

According to an embodiment of the present invention, there is provided a method for preparing data for predicting a truncated image.

Referring to FIG. 1, FIG. 1 is a flowchart of a method 10 for preparing data for predicting a truncated image, according to an exemplary embodiment of the present invention. The method 10 may include step 200 .

As shown in FIG. 1 , in step 200 (virtual simulation step), virtual simulation is performed on the image data to simultaneously obtain a virtual distorted image with data truncation and a virtual refined standard image without data truncation.

The method 10 is to prepare data for predicting a truncated image, for example, preparing data for predicting a truncated image based on an AI method. In this way, two kinds of datasets need to be prepared for AI network learning: one is the input dataset (i.e., distorted images caused by truncation); the other is the refined standard dataset (i.e., without any truncation).

However, in actual clinical cases, since data truncation has occurred (for example, the target object is out of the scanning area), that is, the truncated part of the data has been lost, so the precise standard data cannot be obtained. However, the method 10 for preparing data for a predicted truncated image according to an exemplary embodiment of the present invention can simulate the above-mentioned data truncation in an intelligent manner, thereby simultaneously obtaining a distorted image and a refined standard image.

First Embodiment of Virtual Simulation

Referring to FIG. 2 , in the first embodiment according to the present invention, the above-mentioned step 200 may further include the following sub-steps 210 to 216 .

Specifically, in sub-step 210, an original image without data truncation is received. The original image is usually a sliced image, but it is not excluded that projection images are sometimes provided. However, even if a projection image is provided, it can be converted into a slice image by back-projection processing.

Referring to FIG. 3 , in sub-step 212 , the part (gray-white part) in the original image corresponding to the target object is virtually shifted (offset) to partially move out of the scanning area, thereby obtaining a virtual refined standard image. It should be noted that in this process, the original slice image itself remains unchanged. The scanning field itself is a circle with the focal point as the center, which can be expressed as, for example, DFOV50, where "50" is the diameter of the scanning field, and its unit is cm. When the scanning field is DFOV50, it is partially moved out of the scanning field, that is, the edge of DFOV50 is partially swept.

Next, as shown in FIG. 2 , in sub-step 214 , a simulated scan is performed on the virtual fine standard image and virtual data acquisition is performed to generate virtual truncated data. As mentioned above, the virtual refined standard image obtained in sub-step 212, that is, the part of the image corresponding to the target object in the original image has moved out of the scanning area (swiped over the edge of the DFOV50), and this part of the image will be It will not be scanned, that is, virtual data truncation occurs. Through sub-step 214, virtual truncated data can be obtained, which is projection data.

Finally, in sub-step 216, image reconstruction processing is performed on the virtual truncated data to obtain a virtual distorted image. The image reconstruction processing may specifically be back-projection processing. The resulting virtual distorted image is a slice image.

It can be seen from Fig. 3 that, compared with the original slice image, only the part of the image (gray-white part) corresponding to the target object is shifted in the final virtual fine-standard image, but the gray-white part is in the virtual fine-standard image. is complete without any truncation. Compared with the original sliced image, the finally obtained virtual distorted image is not only the part of the image corresponding to the target object (gray-white part) that has been translated, but also the data of the gray-white part in the virtual distorted image is obviously different. Complete, which has had data truncation. As a result, only through a virtual simulation process, the input data and precise standard data required for AI network learning can be simultaneously obtained from sliced images without data truncation.

Second Embodiment of Virtual Simulation

The virtual simulation process according to the present invention can also be implemented in another way.

Specifically, as shown in FIG. 4 , in the second embodiment according to the present invention, the above-mentioned step 200 may further include the following sub-steps 220 to 226 .

In sub-step 220, a raw image without data truncation is received, the raw image being used as a virtual refined standard image. As mentioned above, the original image is usually a sliced image, but it is not excluded that a projection image is sometimes provided, and in this case, it can be converted into a sliced image by back-projection processing. It can be seen from FIG. 5 that the virtual fine standard image not only has complete data, but also is within the range of the scanning domain (eg, DFOV50).

Next, in sub-step 222, the original image (that is, the virtual fine-standard image) is kept in the scanning domain, and the original image is subjected to orthographic processing to obtain the sinogram of the original image. As mentioned above, the original image does not contain any truncation, and the data in it is complete. Therefore, through the forward projection processing, complete projection data can be obtained from the original slice image, that is, complete sinogram data can be obtained, See Figure 5 for details.

Then, in sub-step 224, the left and right channels of the sinogram are cut off and padded with padding values to generate virtual truncated data. As shown in Figure 5, the original left and right channels in the sinogram are replaced with padding values, thereby simulating truncation.

Regarding the above "padding value", it is padding (padding), and its attribute defines the space between the element border and the element content. The padding shorthand property sets all padding properties in one declaration. Sets all current or specified element padding properties. This property can have 1 to 4 values. Using the padding attribute alone is to set the padding attribute of the element in a declaration. The abbreviated padding attribute can also be used. Once a value is changed, the distance corresponding to the padding will change.

The padding value involved in the present invention is the padding attribute value. It is better to select the edge padding attribute value, that is, select the edge padding value (for example, the value of the outermost channel of the untruncated part) to fill the original left channel in the sinogram. and part of the right channel. However, in actual operation, other padding values (eg, 0) can also be selected for padding according to requirements.

Finally, in sub-step 226, image reconstruction processing is performed on the virtual truncated data to obtain a virtual distorted image. Similar to the first embodiment of the present invention, the image reconstruction process may specifically be a back-projection process, and the obtained virtual distorted image is a slice image. As shown in FIG. 5 , compared with the original sliced image, the finally obtained virtual distorted image has obviously incomplete data, especially the part of the image (gray part) corresponding to the target object, that is, data truncation occurs.

Therefore, a method different from the above-mentioned first embodiment is adopted to simulate data truncation, and in this process, input data (distorted image) and fine standard data (fine standard image) required for AI network learning are obtained at the same time.

Although two embodiments of virtual simulation are described above, it does not mean that the present invention can only simulate data truncation in these two ways. The method for preparing data for prediction of a truncated image according to the present invention can also use other ways to simulate data truncation virtually, so as to simultaneously obtain input data (distorted image) and fine standard data (fine standard image) required for AI network learning ).

According to the virtual simulation process of the present invention, the problem of not being able to obtain precise standard data in the traditional technology is completely solved, and sufficient input data can be obtained very quickly and conveniently.

Optionally, before step 200, the method 10 for preparing data for predicting a truncated image according to an exemplary embodiment of the present invention may further include step 100 (adaptive classification step), see FIG. 6 for details.

Step 100 performs adaptive classification on the collected image data according to predefined features. The image data may be a projection image or a slice image. If it is a projection image, it can be converted into a slice image by conventional back-projection processing. Step 200 further performs virtual simulation on the image data classified in step 100.

Through the above step 100, the method 10 can intelligently classify the image data into different categories before performing virtual simulation on the image data.

Referring to FIG. 7 , in the above step 100, the predefined features may include image data types. Specifically, the image data type may be the anatomical part of the patient. Further, the predefined features may further include possibilities corresponding to the image data types.

For example, the aforementioned anatomical sites may include shoulders, chest, abdomen, pelvis, and phantoms. The models described here can usually be phantoms, but sometimes water models.

In the clinic, different users (e.g. hospitals, patients or research institutes, etc.) will have different patient datasets, so they may need to provide different input datasets for their AI network training. The adaptive classification step 100 according to the example of the present invention can configure the data input, such as the data type, the possibility of the data, etc., which will greatly improve the capabilities of the AI network.

Returning to the method shown in FIG. 6 , when the virtual simulation step 200 is performed, there will always be a non-truncated (no data truncated), complete slice image as the original simulation input. This sliced image comes from a dataset that contains all sliced images of the human body from head to toe. This huge dataset can be a dataset collected by various clinical medical institutions (such as hospitals) and published to the public, or it can be created by users Data set collected by yourself. The adaptive classification step 100 according to the example of the present invention is to perform adaptive classification on the huge data set based on which part of the image the user needs to use.

For example, referring to Figure 7, among the patients in a certain hospital, there may be more shoulder patients, so more patients are irradiated with CT because of shoulder patients, so the obtained slice image is the shoulder part. The probability is relatively high, for example, there is a 40% probability; the next most are pelvic patients, so the patients who are irradiated with CT due to pelvic diseases are more frequently, and correspondingly, the obtained slice image is the pelvic part. The probability is slightly smaller than that of the shoulder, for example, there is a 30% probability; then there are more patients in the chest and abdomen, so the patients who are irradiated with CT for chest and abdomen patients are relatively less likely If it is reduced, correspondingly, the obtained slice images are less likely to be the chest and abdomen, for example, both are 20%. At the same time, there are patients with special models installed in the hospital, so the possibility that the obtained slice images are special models should be considered, such as 10%. According to the patient distribution in the above-mentioned hospitals, special data types (parts of the human body) can be classified and their corresponding possibilities can be optimized. Specifically, for example, the following two characteristics can be pre-defined for data classification:

Feature 1: Special patient parts (as shown in Figure 2, shoulders, chest, abdomen, pelvis, special models);

Feature 2: Special possibility (as shown in Figure 2, 40%, 20%, 20%, 30%, 10%, corresponding to the above-mentioned respective patient parts).

Based on the above two features, the huge image dataset collected by the hospital can be adaptively classified. In this way, the input datasets that are fed into the AI learning network can be selected in a targeted manner.

For another example, for some specialized hospitals, such as chest hospitals, almost all patients should be related to the chest, then the "special patient part" of the above feature 1 can be set to "chest", and the above feature 2 can be set as "chest". The "Special Likelihood" is set to "100%". For another example, even in a general hospital, if the obtained slice images are from a specialized department (for example, neurosurgery), the "special patient part" of the above feature 1 can be set to "brain". part", and the "special possibility" of the above-mentioned feature 2 is set to "100%".

In a word, features can be predefined according to various actual situations and specific needs, and although there are two features defined in the above example, sometimes more features can be defined according to needs, and the features are not only Limited to the type of data and the corresponding possibilities. Furthermore, even the data type is not limited to body parts, but can also be other structures or objects that can be imaged.

It is clear that the above method for preparing data for predicting truncated images can greatly improve the quality of the input data used in the AI learning network because it includes the adaptive classification step as described above.

Further, with the preparation of the above data, the truncated image can be predicted by means of AI. Specifically, reference may be made to FIG. 8 , which shows a flowchart of a method 80 for predicting a truncated image according to an exemplary embodiment of the present invention.

As shown in FIG. 8 , the method 80 for predicting a truncated image according to an exemplary embodiment of the present invention may include step 810 .

In step 810, a truncated image is predicted based on a trained learning network obtained by data training based on a data set consisting of a virtual distorted image and a virtual refined standard image obtained by using the above data preparation method.

The above method 80 obtains a data set for AI network learning through a data preparation method according to an exemplary embodiment of the present invention, in this process not only can adaptive data classification be used to match special patient sets for system calibration of different users. and doctors are used to it, and most importantly, the necessary AI training data (ie, input data and refined standard data) can be obtained at the same time intelligently. The data set thus obtained is undoubtedly very helpful for AI network learning, which can greatly improve the accuracy of predicting truncated images through AI.

In addition, the present invention also provides a system for predicting truncated images.

FIG. 9 shows a system 900 for predicting truncated images according to an exemplary embodiment of the present invention. The system 900 may include a virtual simulation device 910 and a prediction device 920 . The virtual simulation device 910 is used to perform virtual simulation on the image data, so as to obtain a virtual distorted image with data truncation and a virtual fine-standard image without data truncation at the same time. The predicting device 920 is configured to predict the truncated image based on the trained learning network, and the trained learning network is obtained by performing data training based on the data set composed of the virtual distorted image and the virtual refined standard image.

Further, FIG. 10 shows an example of a system for predicting a truncated image according to an exemplary embodiment of the present invention.

As shown in FIG. 10, an exemplary CT system 1000 is configured for predicting truncated images and/or preparing data for predicting truncated images. Specifically, the CT system 1000 is configured to image the target object; if the obtained image has a truncation, then the truncated image is predicted to recover the truncated image data. The above-mentioned target object may be a patient or any other object that needs to be imaged, and in this example, a patient is taken as an example.

In one embodiment, the CT system 1000 includes a gantry 1002 having an X-ray source 1004 and a detector array 1008 formed of a plurality of detector elements 2002 disposed oppositely thereon. X-ray source 1004 is used to project patient-penetrating X-rays 1006 towards detector array 1008, which collects attenuated X-ray beam data that is preprocessed as a projection of the patient's target volume data.

In one embodiment, CT system 1000 includes control mechanism 2008 . The control mechanism 2008 may include an X-ray controller 2010 for providing power and timing signals to the X-ray source 1004 . The control mechanism 2008 may also include a gantry motor controller 2012 for controlling the rotational speed and/or position of the gantry 1002 based on imaging requirements. In addition, the control mechanism 2008 may also include a patient couch controller 2026 for moving the patient couch 2028 to position the patient (shown) in the proper position within the gantry 1002 .

In one embodiment, CT system 1000 further includes a data acquisition system (DAS) 2014 for sampling and digitizing analog data received from detector element 2002 .

In one embodiment, the CT system 100 further includes a computing device 2016 to which the data sampled and digitized by the DAS 2014 will be transmitted to a computer or computing device 2016 for processing. The computing device 2016 can communicate with the operator console 2020 to facilitate the operator's operation, and can also be connected to a display 2032 to facilitate the operator's observation. Additionally, computing device 2016 may also be connected to a picture archiving and communication system (PACS) 2024 .

In one example, computing device 2016 stores data in a storage device, such as computer-readable storage medium 2018 (mass storage in FIG. 10). The computer-readable storage medium 2018 may include a hard disk drive, a floppy disk drive, a compact disk read/write (CD-R/W) drive, a digital versatile disk (DVD) drive, a flash drive, and/or solid state storage, among others.

In addition, computing device 2016 may also be used to provide commands and parameters to DAS 2014, X-ray controller 2010 and one or more of gantry motor controller 2012, patient bed controller 2026 to control system operations, such as data acquisition and/or processing.

The CT system 1000 may further include an image reconstructor 2030 that performs image reconstruction using a suitable image reconstruction method based on the sampled and digitized X-ray data described above. For example, the image reconstructor 2030 may reconstruct an image of the patient's target volume using, for example, filtered back projection (FBP). Although image reconstructor 2030 is illustrated in FIG. 10 as a separate entity, in some embodiments, image reconstructor 2030 may be formed as part of computing device 2016 . Alternatively, image reconstructor 2030 may not be present in CT system 1000; alternatively, computing device 2016 may perform one or more functions of image reconstructor 2030. Additionally, image reconstructor 2030 may be located at a local or remote location and may be operatively connected to CT system 1000 using a wired or wireless network.

In one embodiment, the image reconstructor 2030 stores the reconstructed image in a storage device or computer-readable storage medium 2018 . Alternatively, the image reconstructor 2030 transmits the reconstructed image to the computing device 2016 to generate patient information for diagnosis.

In one embodiment, the CT system 1000 further includes a truncated image prediction device, which can receive the truncated image from the above-mentioned computing device 2016 or the image reconstructor 2030, and predict and/or restore the truncated image, the above-mentioned truncated image prediction device May be formed as part of computing device 2016 or image reconstructor 2030 (part of computing device 2016 in Figure 10).

In one embodiment, CT system 1000 may further include data preparation means for preparing data for truncated image prediction, which data preparation means may also be formed as part of computing device 2016 or image reconstructor 2030 (computing in FIG. 10 ). part of Device 2016).

It should be noted that various methods and processes further described in this specification may be stored in the computer-readable storage medium of the computing device 2016 and/or the image reconstructor 2030 in the CT system 1000 in the form of executable instructions. For example, the truncated image prediction apparatus, the data preparation apparatus may include executable instructions of a computer-readable storage medium, and the methods described in this specification may be employed to predict the truncated image/prepare data.

Some exemplary embodiments have been described above. It should be understood, however, that various modifications may be made to the above-described exemplary embodiments without departing from the spirit and scope of the present invention. For example, if the described techniques are performed in a different order and/or if components in the described systems, architectures, devices, or circuits are combined in different ways and/or are replaced or supplemented by additional components or their equivalents, Appropriate results may be achieved, and accordingly, other embodiments of these modifications also fall within the protection scope of the claims.

Claims (14)

1. A method for preparing data for predicting a truncated image, comprising the steps of: The virtual simulation step is used to perform virtual simulation on the image data, so as to obtain a virtual distorted image with data truncation and a virtual fine standard image without data truncation at the same time. 2. The method according to claim 1, characterized in that, before the virtual simulation step, the method further comprises an adaptive classification step, wherein the adaptive classification step is used to classify the collected images according to predefined characteristics The data is adaptively classified, and the virtual simulation step is used for virtual simulation of the classified image data. 3. The method of claim 2, wherein the predefined characteristic comprises an image data type. 4. The method of claim 3, wherein the predefined characteristics further include a possibility corresponding to the image data type. 5. The method of claim 3, wherein the image data type is a patient's anatomy. 6. The method of claim 1, wherein the virtual simulation step further comprises: Receive raw images without data truncation; virtual translation (offset) part of the original image corresponding to the target object to partially move out of the scanning field, thereby obtaining a virtual refined standard image; performing a simulated scan and virtual data acquisition on the virtual fine-standard image to generate virtual truncated data; and Perform image reconstruction processing on the virtual truncated data to obtain a virtual distorted image. 7. The method of claim 1, wherein the virtual simulation step further comprises: receiving a raw image without data truncation, the raw image being used as a virtual refined standard image; Keeping the original image in the scanning domain, and performing orthographic processing on the original image to obtain a sinogram of the original image; cutting off the left and right channels of the sinogram and padding with padding values to generate virtual truncated data; and Perform image reconstruction processing on the virtual truncated data to obtain a virtual distorted image. 8. The method of claim 7, wherein the padding value is an edge padding value. 9. A method for predicting a truncated image, comprising the steps of: The truncated image is predicted based on a trained learning network based on a data set consisting of a virtual distorted image and a virtual refined standard image obtained by using the method according to any one of claims 1-8. obtained by training. 10. A system for predicting a truncated image, comprising: a virtual simulation device for performing virtual simulation on the image data to simultaneously obtain a virtual distorted image with data truncation and a virtual refined standard image without data truncation; and, The prediction device is used for predicting a truncated image based on a trained learning network obtained by data training based on a data set formed by the virtual distorted image and the virtual refined standard image. 11. The system according to claim 10, further comprising an adaptive classifier for adaptively classifying the collected image data according to predefined characteristics, and the virtual simulation device is used for classifying the collected image data. virtual simulation of the image data. 12. The system of claim 10, wherein the virtual simulation device is further configured to: Receive raw images without data truncation; virtual translation (offset) part of the original image corresponding to the target object to partially move out of the scanning field, thereby obtaining a virtual refined standard image; performing a simulated scan and virtual data acquisition on the virtual fine-standard image to generate virtual truncated data; and Perform image reconstruction processing on the virtual truncated data to obtain a virtual distorted image. 13. The system of claim 10, wherein the virtual simulation device is further configured to: receiving a raw image without data truncation, the raw image being used as a virtual refined standard image; Keeping the original image in the scanning domain, and performing orthographic processing on the original image to obtain a sinogram of the original image; cutting off the left and right channels of the sinogram and padding with padding values to generate virtual truncated data; and Perform image reconstruction processing on the virtual truncated data to obtain a virtual distorted image. 14. A computer-readable storage medium having encoded instructions recorded thereon which, when executed, perform the method of any of claims 1-9.

Images