A micro-dose CT system and a data acquisition method

Abstract

The present application discloses a micro-dose CT system and a data acquisition method. The data acquisition method includes the following steps: predicting the photon starvation state of the target during tomographic scanning; based on the prediction result, determining whether to generate an exposure enable signal and a collection enable signal when a collection trigger signal is generated at the current tomographic scanning angle, selectively skipping the tomographic scanning angles with photon starvation state, and cyclically triggering synchronous exposure and collection according to the exposure enable signal and the collection enable signal. The data acquisition method of the micro-dose CT system of the present application predicts photon starvation, dynamically adjusts the sampling angle according to the prediction result, avoids data acquisition in the photon starvation area, and at the same time, combines pulsed exposure technology to achieve synchronous exposure and collection, improves the image quality on the premise of maintaining a low radiation dose, and further ensures the accuracy of diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of medical devices, and particularly to a micro-dose CT system and a data acquisition method thereof. Background Art

[0002] Computed tomography (CT) is an important medical diagnostic tool, but its radiation dose has always been a concern. Traditional CT scanning methods usually use a relatively high radiation dose to ensure image quality, but this increases the radiation risk to patients. Micro-dose CT imaging technology aims to reduce the radiation dose to the lowest possible level to minimize the radiation risk to patients. However, in micro-dose CT imaging, the photon starvation phenomenon is more serious, resulting in increased image noise, more artifacts, and reduced contrast, which seriously affects the accuracy of diagnosis. Summary of the Invention

[0003] The purpose of the present invention is to provide a micro-dose CT system and a data acquisition method thereof, so as to solve the problem that in the prior art, when reducing the radiation dose in micro-dose CT imaging, the photon starvation phenomenon is aggravated, resulting in increased image noise, more artifacts, and reduced contrast, which seriously affects the accuracy of diagnosis.

[0004] To solve the above technical problems, the technical solution adopted by the present invention is:

[0005] A data acquisition method for a micro-dose CT system includes the following steps:

[0006] Predict the photon starvation state of the target in tomographic scanning;

[0007] Based on the prediction result, determine whether to generate an exposure enable signal and a data acquisition enable signal when generating a data acquisition trigger signal at the current tomographic scanning angle, selectively skip the tomographic scanning angles with the photon starvation state, and cyclically trigger synchronous exposure and data acquisition according to the exposure enable signal and the data acquisition enable signal.

[0008] Further, the photon starvation state is predicted based on a photon starvation prediction algorithm, and the photon starvation prediction algorithm includes:

[0009] Predict by analyzing the attenuation characteristics of the scout view. Perform a scout view scan on the target, and combine the attenuation characteristics of the target scan area obtained from the scout view scan to predict the area where photon starvation occurs in tomographic scanning;

[0010] Or / and, predict by real-time data analysis. Statistically analyze the total number m of pixels with photon starvation in a single data acquisition during the exposure scan of the target, and predict the change trend of the total number m of pixels with photon starvation in consecutive multiple data acquisitions during tomographic scanning of the target;

[0011] Or / and, predict the probability of photon starvation of the target in the scan through an AI model.

[0012] Further, define the photon starvation state as K, the photon starvation index predicted by positioning image attenuation characteristic analysis as k1, the photon starvation index predicted by real-time data analysis as k2, and the probability of photon starvation predicted by the AI model as k3;

[0013] When independently predicting the photon starvation state K through positioning image attenuation characteristic analysis, K = k1, and the threshold of K is 0;

[0014] When independently predicting the photon starvation state K through real-time data analysis, K = k2, and the threshold of K is 0;

[0015] When independently predicting the photon starvation state K through the AI model, K = k3, and the threshold of K is k0;

[0016] When jointly predicting the photon starvation state K through positioning image attenuation characteristic analysis and real-time data analysis, K = k1 + k2, and the threshold of K is 0;

[0017] When jointly predicting the photon starvation state K through positioning image attenuation characteristic analysis and the AI model, K = k1 + k3, and the threshold of K is k0;

[0018] When jointly predicting the photon starvation state K through real-time data analysis and the AI model, K = k2 + k3, and the threshold of K is k0;

[0019] When jointly predicting the photon starvation state K through positioning image attenuation characteristic analysis, real-time data analysis and the AI model, K = k1 + k2 + k3, and the threshold of K is k0;

[0020] Among them, k0 is the photon starvation determination threshold in the AI model prediction.

[0021] Further, based on the angle selection algorithm, determine whether to generate an exposure enable signal and a collection enable signal when a collection trigger signal is generated at the current tomographic scan angle, selectively skip the tomographic scan angles with photon starvation states, and cycle to trigger synchronous exposure and collection according to the exposure enable signal and the collection enable signal.

[0022] Further, based on the angle selection algorithm, determine whether to generate an exposure enable signal and a collection enable signal when a collection trigger signal is generated at the current tomographic scan angle, selectively skip the tomographic scan angles with photon starvation states, and cycle to trigger synchronous exposure and collection according to the exposure enable signal and the collection enable signal, specifically including:

[0023] Step 1: Define the acquisition angle as A and the sparse acquisition angle period as B, where B = 2π / D and D is the total number of sparse acquisition angles;

[0024] Step 2: Start the gantry rotation;

[0025] Step 3: Real-time detect the gantry rotation angle and generate scan control signals, including scan start and end signals. According to the acquisition angle period B, insert acquisition trigger signals between the scan start and end signals. The angular interval between adjacent acquisition trigger signals is B. At the start of the first acquisition trigger signal, trigger a synchronous exposure and acquisition;

[0026] Step 4: When each acquisition is completed, predict the K value for the next acquisition;

[0027] If the K value exceeds the threshold, activate the photon starvation state, calculate the sum W1 of the un-sampled angles. When W1 ≥ B, update the calculation of the K value every A angles. When W1 = xB (where x is a positive integer), trigger a synchronous exposure and acquisition;

[0028] If the K value is below the threshold, turn off the photon starvation state and trigger a synchronous exposure and acquisition. Trigger a synchronous exposure and acquisition after the sum W2 of the acquired angles is yB (where y is a positive integer);

[0029] Loop the above steps.

[0030] Furthermore, trigger the synchronous exposure and acquisition according to the exposure enable signal and the acquisition enable signal. Specifically:

[0031] The exposure enable signal and the acquisition enable signal are generated simultaneously and have equal duration;

[0032] Or, the exposure enable signal and the acquisition enable signal are generated successively, and the exposure enable signal at least satisfies that the rising edge of the exposure enable start is generated slightly earlier than the start time of the acquisition enable signal, and the falling edge of the exposure enable end is generated slightly later than the end time of the acquisition enable signal.

[0033] Furthermore, perform a scout scan on the target, and predict the area where photon starvation occurs in the tomographic scan in combination with the attenuation characteristics of the target scan area obtained from the scout scan. Specifically include:

[0034] Acquire the forward scout image or / and the lateral scout image orthogonal to the target, and define the forward scout image as the 0-degree scout image and the lateral scout image as the 90-degree scout image;

[0035] Obtain the attenuation characteristics of the target scan area at 0 degrees and 90 degrees, and calculate the orthogonal attenuation amounts a0 and a90 of each Z coordinate corresponding tomographic plane;

[0036] Establish the attenuation distribution curve of the fault plane corresponding to each Z coordinate, and obtain the estimated attenuation a(θ) of the target scanning area at each scanning angle by combining the shape of the fault plane in the target scanning area;

[0037] Set the attenuation threshold for photon starvation as bs, and combine the estimated attenuation a(θ) to predict the real-time attenuation b(θ) of the target scanning area at each scanning angle during the scan. When b(θ) < bs, set the photon starvation index k1 = 0, otherwise k1 = 1.

[0038] Furthermore, the calculation formulas for the attenuation a0 and a90 are as follows:

[0039] ;

[0040] ;

[0041] where M and N are the sizes of the pixel matrix of the detector, is the size of each pixel, is the offset of each pixel corresponding to the 0-degree localization image, is the offset of each pixel corresponding to the 90-degree localization image;

[0042] The calculation formula for the real-time attenuation b(θ) is as follows:

[0043] ;

[0044] where kVt is the tube voltage for the localization image scan, mAt is the tube current for the localization image scan, kV(θ) is the tube voltage for the tomographic scan, mA(θ) is the tube current for the tomographic scan, and c is the tube voltage proportionality coefficient.

[0045] Furthermore, count the total number m of pixels that experience photon starvation in a single acquisition during the exposure scan of the target, and predict the change trend of the total number m of pixels that experience photon starvation in multiple consecutive acquisitions during the tomographic scan of the target, specifically including:

[0046] Collect non-exposure data at V angles within 360 degrees of rotation, where V ≥ 100, calculate the expected value μ and standard deviation σ of each pixel at the background noise level, and set the photon starvation judgment threshold THR based on the expected value μ and standard deviation σ;

[0047] Obtain real-time scan data through pulsed scanning and perform pixel-level difference calculations. Compare all pixels with the photon starvation judgment threshold THR to obtain P comparison results, where P ≤ M × N. Set the total number of pixels below the photon starvation judgment threshold THR as m, where m ≤ P, where P is the total number of detector pixels participating in the photon starvation prediction, and M and N are the sizes of the pixel matrix of the detector;

[0048] Store the last three m values: mn , m n-1 , m n-2 , calculate the acceleration a of the change rate of the m value, a = (m n - m n-1 ) - (m n-1 - m n-2 ), predict the change amount in the next cycle = (m n - m n-1 ) + a, set the reference value mo of the total number of pixels with photon starvation and the reference value of the cycle change amount , when m > mo and , k2 = 1, otherwise k2 = 0.

[0049] Furthermore, predict the probability of photon starvation of the target during scanning through the AI model. Specifically:

[0050] The AI model network adopts a multi-modal convolutional neural network;

[0051] Input data:

[0052] The target localization image and scanning parameter information;

[0053] Or, the data of real-time scanning and scanning parameter information;

[0054] Or, the target localization image, scanning parameter information and the data of real-time scanning;

[0055] Output data: The photon starvation probability k3, and the range of k3 is 0 to 1.

[0056] Furthermore, the AI model network structure includes:

[0057] The input layer, used to receive multi-modal data, and the multi-modal data includes images, attenuation curves and scanning parameters;

[0058] The CNN branch, used to process anatomical structure images and extract spatial features;

[0059] The fully connected branch, used to process attenuation curves and scanning parameters and extract physical features;

[0060] The feature fusion layer, used to splice multi-modal features and input them into the fully connected layer;

[0061] The output layer, generating the photon starvation probability k3 based on the Sigmoid activation function.

[0062] Furthermore, it also includes reconstructing the collected data to obtain a CT image, and the reconstruction of the CT image adopts an iterative reconstruction algorithm or an AI reconstruction algorithm.

[0063] The present invention also provides a micro-dose CT system, which is based on the above data acquisition method and includes:

[0064] A gantry;

[0065] An X-ray tube, which is arranged on the gantry and includes a cathode, an anode, and a grid control electrode arranged between the cathode and the anode;

[0066] A high-voltage generator, which is arranged on the gantry and is connected to the X-ray tube for providing a working voltage and a pulsed grid voltage to the X-ray tube;

[0067] A detector, which is arranged on the gantry and is connected to the X-ray tube for receiving X-rays and converting them into electrical signals;

[0068] A data acquisition unit, which is signal-connected to the detector for acquiring, processing, and storing the electrical signals output by the detector;

[0069] An image reconstruction unit, which is signal-connected to the data acquisition unit for reconstructing the data acquired by the data acquisition unit to generate a CT image;

[0070] A control unit, which is arranged on the gantry and is respectively signal-connected to the X-ray tube, the high-voltage generator, the detector, the data acquisition unit, and the image reconstruction unit.

[0071] Due to the application of the above technical solution, the beneficial effects of the present application compared with the prior art are as follows:

[0072] The data acquisition method of the micro-dose CT system of the present application predicts photon starvation and dynamically adjusts the sampling angle according to the prediction result, avoiding data acquisition in the photon starvation area. At the same time, combined with the pulsed exposure technology, synchronous exposure and acquisition are realized, improving the image quality on the premise of maintaining a low radiation dose, and thus ensuring the accuracy of diagnosis. Description of the Drawings

[0073] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0074] Figure 1 It is a flowchart of the angle selection algorithm in Embodiment 1 of the present invention;

[0075] Figure 2 It is a timing diagram of the photon starvation state and the acquisition enable signal in Embodiment 1 of the present invention;

[0076] Figure 3 This is a schematic structural diagram of the X-ray tube in Embodiment 2 of the present invention.

[0077] Explanation of the reference numerals in the drawings:

[0078] 1 - Cathode; 2 - Anode; 3 - Grid control electrode. Detailed implementation manners

[0079] In order to enable those skilled in the art of the present technology to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0080] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so as to describe the embodiments of the present application here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0081] In the present application, the orientation or positional relationship indicated by the terms "upper", "lower", "left", "right", "front", "rear", "top", "bottom", "inner", "outer", "middle", "vertical", "horizontal", "lateral", "longitudinal", etc. is based on the orientation or positional relationship shown in the drawings. These terms are mainly used to better describe the present invention and its embodiments, and are not used to limit that the indicated devices, elements or components must have a specific orientation or be constructed and operated in a specific orientation.

[0082] Moreover, in addition to being able to be used to represent the orientation or positional relationship, some of the above-mentioned terms may also be used to represent other meanings. For example, the term "upper" may also be used to represent a certain attachment relationship or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in the present invention can be understood according to specific circumstances.

[0083] In addition, the terms "installed", "set up", "equipped with", "connected", "linked", "socketed" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, or there can be internal communication between two devices, components or parts. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0084] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will detail this application with reference to the drawings and in conjunction with the embodiments.

[0085] Embodiment 1

[0086] This embodiment provides a data acquisition method for a micro-dose CT system, which can be applied to various CT scanning scenarios, such as chest CT screening, pediatric CT examination, etc., and is particularly suitable for occasions where radiation dose needs to be controlled. This application does not limit this. The data acquisition method includes the following steps:

[0087] Predict the photon starvation state of the target in tomographic scanning;

[0088] Based on the prediction result, determine whether to generate an exposure enable signal and a data acquisition enable signal when generating a data acquisition trigger signal at the current tomographic scanning angle, selectively skip the tomographic scanning angles with photon starvation state, and cycle to trigger synchronous exposure and data acquisition according to the exposure enable signal and the data acquisition enable signal.

[0089] In this embodiment, the photon starvation state is predicted based on a photon starvation prediction algorithm. The core purpose of the photon starvation prediction algorithm is to optimize the exposure and data acquisition strategies during CT scanning by predicting the occurrence of photon starvation phenomena during image acquisition, reduce the radiation dose, and improve the image quality.

[0090] In this embodiment, the photon starvation prediction algorithm includes predicting through positioning image attenuation characteristic analysis. Perform a positioning image scan on the target, and combine the attenuation characteristics of the target scanning area obtained from the positioning image scan to predict the area where photon starvation occurs in the tomographic scanning of the target; or / and, predict through real-time data analysis. Count the total number of pixels m that experience photon starvation in a single acquisition during the exposure scan of the target, and predict the change trend of the total number of pixels m that experience photon starvation in consecutive multiple acquisitions during the tomographic scanning of the target; or / and predict the probability of photon starvation occurring in the target during scanning through an AI model.

[0091] It is worth noting that the above three photon starvation prediction algorithms can be used to perform photon starvation prediction independently, or can be combined in pairs or all three together for photon prediction, and this application does not make specific limitations on this. Of course, in other implementation schemes, other photon starvation prediction algorithms can also be used, and this application does not elaborate on them here.

[0092] Specifically, the photon starvation state is defined as K, the photon starvation index predicted by the positioning image attenuation characteristic analysis is k1, the photon starvation index predicted by real-time data analysis is k2, and the photon starvation probability predicted by the AI model is k3. The photon starvation state prediction is as follows:

[0093] When the photon starvation state K is predicted independently through the positioning image attenuation characteristic analysis, K=k1, and the threshold of K is 0; when the photon starvation state K is predicted independently through real-time data analysis, K=k2, and the threshold of K is 0; when the photon starvation state K is predicted independently through the AI model, K=k3, and the threshold of K is k0; when the photon starvation state K is predicted jointly through the positioning image attenuation characteristic analysis and real-time data analysis, K=k1+k2, and the threshold of K is 0; when the photon starvation state K is predicted jointly through the positioning image attenuation characteristic analysis and the AI model, K=k1+k3, and the threshold of K is k0; when the photon starvation state K is predicted jointly through the real-time data analysis and the AI model, K=k2+k3, and the threshold of K is k0; when the photon starvation state K is predicted jointly through the positioning image attenuation characteristic analysis, the real-time data analysis and the AI model, K=k1+k2+k3, and the threshold of K is k0.

[0094] It should be noted that the above k0 is the photon starvation judgment threshold in the AI model prediction, and the K threshold of 0 means that when K=0, the CT system is in a normal state. In some optional implementations, k0 is usually greater than a certain value, such as 0.6, or a number of thresholds between 0.6 and 1 are set as judgment conditions of different levels, such as 0.6, 0.7, 0.8, and 0.9. The larger the value, the more relaxed the probability of allowing the photon starvation state to occur.

[0095] In this embodiment, a locator image scan is performed on the target, and the attenuation characteristics of the target scanning area obtained by the locator image scan are combined to predict the photon-starved area of the target in the tomographic scan, specifically including:

[0096] Collecting a forward positioning image or / and a lateral positioning image orthogonal to the target, wherein the forward positioning image is defined as a 0-degree positioning image and the lateral positioning image is defined as a 90-degree positioning image;

[0097] Obtain the attenuation characteristics of the target scanning area at 0 degrees and 90 degrees, and calculate the orthogonal attenuation a0 and a90 of each Z coordinate corresponding to the fault plane;

[0098] Establish the attenuation distribution curve of the fault plane corresponding to each Z coordinate, and obtain the estimated attenuation a(θ) of the target scanning area at each scanning angle by combining the shape of the fault plane in the target scanning area. For example, if the elliptical shape is used to simulate the shape of the human body's fault plane, the calculation formula for a(θ) is as follows:

[0099] ;

[0100] Set the attenuation threshold for photon starvation as bs, and predict the real-time attenuation b(θ) of the target scanning area at each scanning angle during the scan by combining the estimated attenuation a(θ). When b(θ) < bs, set the photon starvation index k1 = 0, otherwise k1 = 1.

[0101] In an alternative embodiment, collect a forward positioning image and a lateral positioning image orthogonal to the target. Then the calculation formulas for the attenuation amounts a0 and a90 are as follows:

[0102] ;

[0103] ;

[0104] Where M and N are the sizes of the pixel matrix of the detector, is the size of each pixel, is the offset of each pixel corresponding to the 0-degree positioning image, is the offset of each pixel corresponding to the 90-degree positioning image;

[0105] The calculation formula for the real-time attenuation b(θ) is as follows:

[0106] ;

[0107] Where kVt is the tube voltage for the positioning image scan, mAt is the tube current for the positioning image scan, kV(θ) is the tube voltage for the tomographic scan, mA(θ) is the tube current for the tomographic scan, and c is the tube voltage proportionality coefficient. The empirical value of c is generally 1.7.

[0108] In another alternative embodiment, collect a forward positioning image or a lateral positioning image orthogonal to the target, and obtain the attenuation characteristics of the other uncollected positioning image by accumulating and normalizing all the pixels in the same Z direction. This is a conventional method and will not be elaborated here.

[0109] In some alternative embodiments, count the total number m of pixels with photon starvation in a single acquisition during the exposure scan of the target, and predict the change trend of the total number m of pixels with photon starvation in consecutive multiple acquisitions during the tomographic scan of the target. Specifically, it includes:

[0110] Collect non-exposure data by rotating 360 degrees to acquire V angles, where V ≥ 100. Calculate the expected value μ and standard deviation σ of each pixel at the background noise level, and set the photon starvation judgment threshold THR based on the expected value μ and standard deviation σ.

[0111] Perform pulse scanning to obtain real-time scan data and conduct pixel-level difference calculation. Compare all pixels with the photon starvation judgment threshold THR to obtain P comparison results, where P ≤ M × N. Set the total number of pixels below the photon starvation judgment threshold THR as m, where m ≤ P. Here, P is the total number of detector pixels participating in photon starvation prediction, and M and N are the pixel matrix sizes of the detector.

[0112] Store the most recent three m values: m n , m n-1 , m n-2 , calculate the acceleration a of the change rate of m, a = (m n - m n-1 ) - (m n-1 - m n-2 ), predict the change amount in the next cycle = (m n - m n-1 ) + a, set the baseline value mo of the total number of pixels with photon starvation and the baseline value of the cycle change amount. When m > mo and , k2 = 1, otherwise k2 = 0.

[0113] It should be noted that the above non-exposure data collection step can be selected to be executed at any non-exposed time of the CT system, and it is preferably automatically executed once before each exposure collection.

[0114] In some alternative implementation schemes, predict the probability of photon starvation of the target during scanning through an AI model. Specifically, the AI model network uses a multi-modal convolutional neural network, and the input data is the target localization image and scanning parameter information; or, the input data is the real-time scan data and scanning parameter information; or, the input data is the target localization image, scanning parameter information, and real-time scan data, and the output data is the photon starvation probability k3, where the range of k3 is 0 to 1.

[0115] It should be noted that the attenuation characteristics of the target scanning area are actually derivatives of the target localization image. Usually, the localization image or attenuation characteristics can be directly input. For the AI model, it is usually simplified and the end-to-end localization image is directly input. The scanning parameter information mainly includes the tube current, tube voltage, rotation time, etc. of the tomographic scan. The real-time scan data is mainly the exposure data of the detector at each angle during the tomographic scan process, and the total number of pixels m with photon starvation in the real-time scan is also a derivative of this parameter.

[0116] Specifically, the AI model network structure includes an input layer, a CNN branch, a fully connected branch, a feature fusion layer, and an output layer. The input layer is used to receive multimodal data, which includes images, attenuation curves, and scanning parameters. The CNN branch is used to process anatomical structure images and extract spatial features. The fully connected branch is used to process attenuation curves and scanning parameters and extract physical features. The feature fusion layer is used to splice multimodal features and input them into the fully connected layer. The output layer generates the photon starvation probability k3 based on the Sigmoid activation function. Through the fusion of multimodal data, this network structure combines the spatial features of images, attenuation curves, and physical features of scanning parameters, thereby achieving accurate prediction of the photon starvation state.

[0117] The training method of the above AI model mainly includes two key steps: dataset construction and data augmentation. First, for dataset construction, CT images with complete annotations need to be collected. These images should cover scanning data of different body parts, such as the chest, pelvis, etc., and each image should be accompanied by corresponding annotation information. In addition, scanning technical parameters of the images, such as tube voltage, tube current, attenuation value, scanning intensity, etc., also need to be recorded. These technical parameters can help the model better understand the features and structures of the images. To further improve the model's sensitivity to image quality, it is also necessary to annotate quality problems such as photon starvation for the images, and clarify whether there is a decrease in image quality during the scanning process.

[0118] Data augmentation is an effective technique to improve the generalization ability of the model. By performing operations such as rotation, translation, scaling, mirror flipping, and noise injection on the original images, different scanning conditions and possible image variations can be simulated. This augmentation method helps to expand the diversity of training data, ensuring that the model can maintain high robustness and accuracy when facing various variations and challenges in practical applications, thereby enhancing its adaptability and performance in different clinical scenarios.

[0119] Please combine Figure 1 and Figure 2 , in this embodiment, when judging whether to generate an exposure enable signal and a collection enable signal at the current tomographic scanning angle based on the angle selection algorithm, the tomographic scanning angles with photon starvation states are selectively skipped, and synchronous exposure and collection are triggered cyclically according to the exposure enable signal and the collection enable signal. The specific steps are as follows:

[0120] Step 1: Define the acquisition angle as A, and the sparse acquisition angle period as B, where B = 2π / D, and D is the total number of sparse acquisition angles;

[0121] Step 2: Start the gantry rotation;

[0122] Step 3: Real-time detect the rotation angle of the gantry and generate a scan control signal, including scan start and end signals. According to the acquisition angle period B, insert acquisition trigger signals between the scan start and end signals, with an angular interval of B between adjacent acquisition trigger signals. At the start of the first acquisition trigger signal, trigger a synchronous exposure and acquisition once;

[0123] Step 4: Each time after the acquisition is completed, predict the K value for the next acquisition;

[0124] If the K value exceeds the threshold, activate the photon starvation state, calculate the sum W1 of the un-sampled angles. When W1 ≥ B, update the calculation of the K value every A angles. When W1 = xB, trigger a synchronous exposure and acquisition once, where x is a positive integer;

[0125] If the K value is lower than the threshold, turn off the photon starvation state and trigger a synchronous exposure and acquisition once. After the sum W2 of the acquired angles reaches yB angles, trigger a synchronous exposure and acquisition once, where y is a positive integer;

[0126] Loop the above steps.

[0127] When there is no photon starvation, the system normally performs sparse pulsed exposure and data acquisition, generally uniform angular exposure and acquisition, to achieve micro-dose scanning; when the photon starvation state occurs, the system dynamically determines the angular interval of pulsed exposure and acquisition according to the angle selection algorithm, and its angular interval is no longer uniform and is affected by the angular distribution of photon starvation. Photon starvation prediction is calculated at least once for each originally evenly distributed angle, and the calculation result is used to update the photon starvation state K. At the same time, by forcing a synchronous exposure and acquisition when W1 = xB and W2 = yB, the rationality of the acquisition angle distribution is ensured.

[0128] In this embodiment, triggering synchronous exposure and acquisition according to the exposure enable signal and the acquisition enable signal specifically means:

[0129] The exposure enable signal and the acquisition enable signal are generated simultaneously and have equal durations;

[0130] Alternatively, the exposure enable signal and the acquisition enable signal are generated successively, and the exposure enable signal at least satisfies that the rising edge of the exposure enable start is generated slightly earlier than the start time of the acquisition enable signal, and the falling edge of the exposure enable end is generated slightly later than the end time of the acquisition enable signal. The advantage of generating successively is to make the X-ray exposure stable enough during the duration of the acquisition enable.

[0131] It should be noted that in some alternative embodiments, the exposure enable signal generates the rising edge of the exposure enable start a few microseconds to several hundred microseconds earlier than the acquisition enable signal, and generates the falling edge of the exposure enable end a few microseconds to several hundred microseconds later than the acquisition enable signal ends. Specifically, it can be 1 microsecond, 10 microseconds, 20 microseconds, 100 microseconds, 500 microseconds, 800 microseconds, and the present application does not limit this.

[0132] In some alternative embodiments, the scan start feature and the scan end feature can be characterized by the same period but different duty cycles.

[0133] In some alternative embodiments, the scan start signal is the first acquisition trigger signal, and the scan end signal is the last acquisition trigger signal.

[0134] The data acquisition aspect of the present application further includes reconstructing the acquired data to obtain a CT image, and the reconstruction of the CT image adopts an iterative reconstruction algorithm or an AI reconstruction algorithm.

[0135] In this embodiment, the AI reconstruction algorithm is used for reconstruction. Specifically, the method of FBP-based reconstruction and high-quality restoration reconstruction by neural network is adopted. This kind of AI model belongs to the image domain AI model. Combining the computational efficiency of FBP and the image enhancement ability of the neural network can effectively remove noise and artifacts, retain the details of anatomical structures, improve the quality of low-dose and few-view CT images, meet the clinical real-time processing requirements, and thus improve the accuracy and value of diagnosis.

[0136] Specifically, for the FBP-based reconstruction part:

[0137] 1. Data preprocessing: Obtain projection data, including offset correction, air correction, ray hardening correction, and nonlinear correction.

[0138] 2. FBP filtered backprojection: After obtaining the projection data through the above steps, filter the projection data based on the FBP algorithm, and then reconstruct the image by backprojection. The mathematical model of this process is based on Fourier transform, and the internal structure of the object is deduced by using the information in the projection data. This is a conventional method and will not be elaborated here.

[0139] For the high-quality restoration reconstruction part by neural network:

[0140] Network architecture: An improved U-Net network structure is adopted, which includes an encoder-decoder architecture, and the detailed information of the image is retained through skip connections. Skip connections can help the network make full use of the details in the input image during the reconstruction in the decoding stage.

[0141] Characteristics of the improved U-Net network:

[0142] Multi-scale feature extraction ability: U-Net extracts multi-scale features in images through multi-level convolution and pooling operations, which helps to improve the sensitivity to details of different sizes, especially the recognition of important features such as edges and textures.

[0143] Attention mechanism: enables the network to better focus on important feature regions while ignoring unimportant regions, thus improving the quality of the reconstructed image.

[0144] Processing flow:

[0145] Input: The input to the network is low-quality CT images, which are obtained after FBP reconstruction and usually have problems such as noise and artifacts.

[0146] Network processing: The improved U-Net network processes low-quality CT images, utilizes its deep feature extraction and restoration capabilities, gradually denoises, restores lost details, and enhances the image quality.

[0147] Output: After network processing, the output is a high-quality CT image, in which noise and artifacts are effectively suppressed and details are enhanced.

[0148] Training strategy:

[0149] Use a paired dataset for training, that is, low-quality images and corresponding high-quality reference images. The paired dataset can help the network learn how to convert low-quality images into high-quality images.

[0150] Obtaining high-quality reference images: By performing full-angle FBP reconstruction on the original data of a complete scan, generally about 1000 angles, high-quality reference images can be obtained.

[0151] Generating low-quality images from sparse projection data: Downsample the projection data of the full-angle data to simulate sparse projection data in a real scenario, and then use the FBP algorithm for reconstruction to obtain low-quality images.

[0152] Data augmentation: To improve the generalization ability of the network, data augmentation techniques such as rotation, scaling, and flipping are adopted to enable the model to handle various different image variations.

[0153] Dataset characteristics: This dataset is generated based on a completely real CT scan scenario. These data are preprocessed and a paired dataset is generated through the FBP algorithm. Since actual CT scan data is used, network training can better simulate real reconstruction scenarios, thereby improving the quality of the reconstruction results.

[0154] In some other embodiments, a projection domain AI model is used for CT image reconstruction. Specifically, sparse projection data is used to generate full-angle projection data through an AI model. Commonly used models include Unet, CNN, or their variants, such as adding residual connections, introducing wavelet transforms, etc. The following are the steps of this method:

[0155] 1. Data preprocessing: The preprocessing in this stage is the same as the above FBP data preprocessing method, and the purpose is to obtain sparse projection data. Usually, the sparse projection data comes from 100 angles or fewer projection angles, and these data will be used as the input of the AI model.

[0156] 2. Model processing: The sparse projection data is input into the trained AI model, and the model will output full-angle projection data. The number of full-angle projection data generally reaches 1000 angles or more, aiming to supplement the missing angle information for subsequent image reconstruction.

[0157] 3. Image reconstruction: Using the generated full-angle projection data, the traditional FBP algorithm is applied to reconstruct the CT image. In this way, a relatively complete CT image can be obtained, although the initially input data is sparse.

[0158] 4. Model training: The training method is similar to the training process of other AI models. First, using real complete scan data, sparse projection data is obtained through downsampling, that is, reducing the number of angle samplings, to form a paired data set. These paired data sets are used to train the AI model so that it can accurately predict full-angle projection data from sparse projection data.

[0159] The advantage of this method is that the AI model can learn the laws and structures of image reconstruction, recover complete projection data from limited projection data, thereby improving the scanning efficiency and reducing the dependence on a large amount of projection data.

[0160] In still some other embodiments, an image domain AI model is combined with a projection domain AI model. This application does not make specific limitations on this.

[0161] Embodiment 2

[0162] This embodiment provides a micro-dose CT system. Based on the data acquisition method of Embodiment 1, the micro-dose CT system includes a gantry and an X-ray tube, a control unit, a high-voltage generator, a detector, a data acquisition unit, and an image reconstruction unit provided on the gantry. The control unit is respectively connected to the X-ray tube, the high-voltage generator, the detector, the data acquisition unit, and the image reconstruction unit by signals.

[0163] Specifically, refer to Figure 3, The X-ray tube includes a cathode 1, an anode 2, and a grid control electrode 3 disposed between the cathode 1 and the anode 2. Only when a high voltage is applied between the cathode 1 and the anode 2, electrons are emitted from the cathode 1 and strike the anode 2 to generate X-rays. When a certain voltage is applied to the grid control electrode 3, the electrons will no longer be emitted to the anode 2, and the generation of X-rays is interrupted. By intermittently applying a voltage to the grid control electrode 3, intermittent pulsed exposure can be achieved.

[0164] The high-voltage generator is connected to the X-ray tube and is used to provide the working voltage and the pulsed grid voltage for the X-ray tube. It should be noted that within the duration of the scan control signal, the tube voltage of the X-ray tube is continuously loaded, and at the same time, the voltage of the grid control electrode 3 is kept default loaded.

[0165] The detector is disposed on the gantry and is connected to the X-ray tube, and is used to receive X-rays and convert them into electrical signals.

[0166] The data acquisition unit is signal-connected to the detector and is used to acquire, process, and store the electrical signals output by the detector. Further, preferably, a data acquisition unit with an FPGA is used to realize real-time data analysis, perform real-time parameter calculation on the acquired data for the calculation of the photon starvation index k2, calculate the real-time photon starvation statistic value m, and provide k2 to the control unit.

[0167] The image reconstruction unit is signal-connected to the data acquisition unit and is used to reconstruct the data acquired by the data acquisition unit to generate a CT image. Further, it participates in the calculation of the photon starvation prediction and the angle selection algorithm, calculates the photon starvation index k1, undertakes the prediction of the photon starvation state of the AI model, obtains the m value of the data acquisition unit in real time, obtains the exposure parameters of the control unit in real time, uses the AI model to infer and predict the photon starvation probability k3, and provides the photon starvation state k1 or / and k3 to the control unit.

[0168] The control unit is used to control the operation of the scan gantry, the X-ray tube, the detector, and the data acquisition unit, provide exposure-related parameters such as tube voltage, tube current, angle information, etc., obtain k1, k2, k3 and perform fusion, conduct the final prediction of the photon starvation state, implement the process control of the angle selection algorithm based on the prediction result, generate the exposure and data acquisition synchronization signals according to the angle selection algorithm, and achieve precise pulsed exposure and data acquisition synchronization. It should be noted that the control unit can use a single-chip microcomputer or an embedded processor as the core controller.

[0169] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A data acquisition method for a micro-dose CT system, characterized in that, It includes the following steps: Predict the photon starvation state of the target in tomographic scanning; Based on the prediction result, determine whether to generate an exposure enable signal and a collection enable signal when a collection trigger signal is generated at the current tomographic scanning angle, selectively skip the tomographic scanning angles with photon starvation states, and cycle to trigger synchronous exposure and collection according to the exposure enable signal and the collection enable signal.

2. The data acquisition method of a micro-dose CT system according to claim 1, characterized in that, Predict the photon starvation state based on a photon starvation prediction algorithm, and the photon starvation prediction algorithm includes: Predict through positioning image attenuation characteristic analysis. Perform a positioning image scan on the target, and combine the attenuation characteristics of the target scan area obtained from the positioning image scan to predict the area where the target has photon starvation in tomographic scanning; Or / And, predict through real-time data analysis. Statistically analyze the total number of pixels m with photon starvation in a single collection during the exposure scan of the target, and predict the change trend of the total number of pixels m with photon starvation in multiple consecutive collections of the target during tomographic scanning; Or / And, predict the probability of the target having photon starvation during scanning through an AI model.

3. The data acquisition method of a micro-dose CT system according to claim 2, characterized in that, Define the photon starvation state as K, the photon starvation index predicted through positioning image attenuation characteristic analysis as k1, the photon starvation index predicted through real-time data analysis as k2, and the photon starvation probability predicted through the AI model as k3; When independently predicting the photon starvation state K through positioning image attenuation characteristic analysis, K = k1, and the threshold of K is 0; When independently predicting the photon starvation state K through real-time data analysis, K = k2, and the threshold of K is 0; When independently predicting the photon starvation state K through the AI model, K = k3, and the threshold of K is k0; When jointly predicting the photon starvation state K through positioning image attenuation characteristic analysis and real-time data analysis, K = k1 + k2, and the threshold of K is 0; When jointly predicting the photon starvation state K through positioning image attenuation characteristic analysis and the AI model, K = k1 + k3, and the threshold of K is k0; When jointly predicting the photon starvation state K through real-time data analysis and the AI model, K = k2 + k3, and the threshold of K is k0; When jointly predicting the photon starvation state K through positioning image attenuation characteristic analysis, real-time data analysis and the AI model, K = k1 + k2 + k3, and the threshold of K is k0; Wherein, k0 is the photon starvation determination threshold in the prediction of the AI model.

4. A data acquisition method for a micro-dose CT system according to any one of claims 1 to 3, characterized in that, Based on an angle selection algorithm, determine whether to generate an exposure enable signal and a collection enable signal when a collection trigger signal is generated at the current tomographic scanning angle, selectively skip the tomographic scanning angles with photon starvation states, and cycle to trigger synchronous exposure and collection according to the exposure enable signal and the collection enable signal.

5. The data acquisition method of a micro-dose CT system according to claim 4, wherein, Based on an angle selection algorithm, determine whether to generate an exposure enable signal and a collection enable signal when a collection trigger signal is generated at the current tomographic scanning angle, selectively skip the tomographic scanning angles with photon starvation states, and cycle to trigger synchronous exposure and collection according to the exposure enable signal and the collection enable signal, specifically including: Step 1: Define the collection angle as A, and the sparse collection angle period as B, where B = 2π / D, and D is the total number of sparse collection angles; Step 2: Start the gantry rotation; Step 3: Detect the rotation angle of the gantry in real time and generate scan control signals, including scan start and end signals. According to the acquisition angle period B, insert acquisition trigger signals between the scan start and end signals, with an angular interval of B between adjacent acquisition trigger signals. At the start of the first acquisition trigger signal, trigger a synchronous exposure and acquisition; Step 4: When each acquisition is completed, predict the K value for the next acquisition; If the K value exceeds the threshold, activate the photon starvation state, calculate the sum W1 of the un-sampled angles. When W1≥B, update and calculate the K value every A angles. When W1 = xB, trigger a synchronous exposure and acquisition, where x is a positive integer; If the K value is lower than the threshold, turn off the photon starvation state and trigger a synchronous exposure and acquisition. After the sum W2 of the acquired angles is yB angles, trigger a synchronous exposure and acquisition, where y is a positive integer; Loop the above steps.

6. The data acquisition method of a micro-dose CT system according to claim 5, characterized in that, Trigger synchronous exposure and acquisition according to the exposure enable signal and the acquisition enable signal. Specifically: The exposure enable signal and the acquisition enable signal are generated simultaneously and have equal duration; Alternatively, the exposure enable signal and the acquisition enable signal are generated successively, and the exposure enable signal at least satisfies that the rising edge of the exposure enable start is generated slightly earlier than the start time of the acquisition enable signal, and the falling edge of the exposure enable end is generated slightly later than the end time of the acquisition enable signal.

7. The data acquisition method of a micro-dose CT system according to claim 3, characterized in that Perform a localization image scan on the target, and predict the area where photon starvation occurs in the target during tomographic scanning by combining the attenuation characteristics of the target scan area obtained from the localization image scan. Specifically include: Acquire the forward localization image or / and the lateral localization image orthogonal to the target, define the forward localization image as the 0-degree localization image, and the lateral localization image as the 90-degree localization image; Obtain the attenuation characteristics of the target scan area at 0 degrees and 90 degrees, and calculate the orthogonal attenuation amounts a0 and a90 of each cross-sectional plane corresponding to each Z coordinate; Establish the attenuation distribution curve of each cross-sectional plane corresponding to each Z coordinate, and obtain the estimated attenuation amount a(θ) of the target scan area at each scan angle by combining the cross-sectional shape of the target scan area; Set the attenuation threshold for photon starvation as bs, and combine the estimated attenuation amount a(θ) to predict the real-time attenuation amount b(θ) of the target scan area at each scan angle during scanning. When b(θ) < bs, set the photon starvation index k1 = 0, otherwise k1 = 1.

8. The data acquisition method of a micro-dose CT system according to claim 7, characterized in that, The calculation formulas for the attenuation amounts a0 and a90 are as follows: ； ； where M and N are the sizes of the pixel matrix of the detector, is the size of each pixel, is the offset of each pixel corresponding to the 0-degree localization image, is the offset of each pixel corresponding to the 90-degree localization image; The calculation formula for the real-time attenuation amount b(θ) is as follows: ； Where, kVt is the tube voltage of the localization image scan, mAt is the tube current of the localization image scan, kV(θ) is the tube voltage of the tomographic scan, mA(θ) is the tube current of the tomographic scan, and c is the tube voltage proportionality coefficient.

9. The data acquisition method of a micro-dose CT system according to claim 3, characterized in that, Count the total number m of pixels with photon starvation in a single acquisition during the exposure scan of the target, and predict the change trend of the total number m of pixels with photon starvation in consecutive multiple acquisitions during the tomographic scan of the target. Specifically include: Acquire non-exposure data at V angles within 360 degrees of rotation, V≥100, calculate the expected value μ and standard deviation σ of each pixel at the background noise level, and set the photon starvation judgment threshold THR according to the expected value μ and standard deviation σ; Pulse scanning is used to obtain real-time scan data and perform pixel-level difference calculation. All pixels are compared with the photon starvation judgment threshold THR to obtain P comparison results, where P ≤ M × N. The total number of pixels below the photon starvation judgment threshold THR is set as m, and m ≤ P. Here, P is the total number of detector pixels participating in photon starvation prediction, and M and N are the pixel matrix sizes of the detector. Store the last three m values: m n , m n-1 , m n-2 , calculate the acceleration a of the change rate of the m value, a = (m n - m n-1 ) - (m n-1 - m n-2 ), predict the change amount in the next cycle = (m n - m n-1 ) + a, set the reference value mo of the total number of pixels with photon starvation and the reference value of the cycle change amount , when m > mo and , k2 = 1, otherwise k2 = 0.

10. The data acquisition method of a micro-dose CT system according to claim 3, characterized in that, The probability of photon starvation occurring in the target during scanning is predicted through an AI model. Specifically: The AI model network adopts a multi-modal convolutional neural network. Input data: The target localization image and scan parameter information; Or, the real-time scan data and scan parameter information; Or, the target localization image, scan parameter information, and real-time scan data; Output data: The photon starvation probability k3, and the range of k3 is 0 to 1.

11. The data acquisition method of a micro-dose CT system according to claim 10, characterized in that, The AI model network structure includes: An input layer for receiving multi-modal data, where the multi-modal data includes images, attenuation curves, and scan parameters; A CNN branch for processing anatomical structure images and extracting spatial features; A fully connected branch for processing attenuation curves and scan parameters and extracting physical features; A feature fusion layer for splicing multi-modal features and inputting them into a fully connected layer; An output layer for generating the photon starvation probability k3 based on the Sigmoid activation function.

12. The data acquisition method of a micro-dose CT system according to claim 1, wherein It also includes reconstructing the acquired data to obtain a CT image, and the reconstruction of the CT image adopts an iterative reconstruction algorithm or an AI reconstruction algorithm.

13. A micro-dose CT system, based on the data acquisition method according to any one of claims 1-12, characterized in that, It includes: A gantry; An X-ray tube, which is arranged on the gantry. The X-ray tube includes a cathode, an anode, and a grid control electrode arranged between the cathode and the anode; A high-voltage generator, which is arranged on the gantry and is connected to the X-ray tube for providing a working voltage and a pulsed grid voltage to the X-ray tube; A detector, which is arranged on the gantry and is connected to the X-ray tube for receiving X-rays and converting them into electrical signals; A data acquisition unit, which is signal-connected to the detector for collecting, processing, and storing the electrical signals output by the detector; An image reconstruction unit, which is signal-connected to the data acquisition unit for reconstructing the data obtained by the data acquisition unit to generate a CT image; A control unit, which is arranged on the gantry and is respectively signal-connected to the X-ray tube, the high-voltage generator, the detector, the data acquisition unit, and the image reconstruction unit.

Images