Image reconstruction method and system based on photon counting detector self-supervised noise reduction, medium, program product and terminal

Abstract

According to the image reconstruction method and system based on self-supervised noise reduction of the photon counting detector, the medium, the program product and the terminal provided by the invention, by constructing the noise propagation model of the photon counting CT technology in the material decomposition process, the counting noise prediction precision of the detector is improved based on the neural network; and then a self-supervised noise reduction result is optimized by using the noise propagation model, and the problem of insufficient photon counting CT training data at present is solved. According to the method, a complex modeling process and a time-consuming calculation process are not needed, response laws of different photon counting detectors can be accurately fitted only through a small amount of data, the problem of noise amplification in the material decomposition and image reconstruction process of the PCCT technology is solved, and the image quality and the processing efficiency are greatly improved. The noise reduction method based on deep learning shows extremely high practical value in the field of clinical application, is expected to provide more reliable and high-quality technical support for related work such as medical diagnosis, and promotes further development of clinical practice.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of photon counting CT technology and self-supervised denoising, and particularly relates to an image reconstruction method and system based on self-supervised denoising of a photon counting detector, a medium, a program product and a terminal. BACKGROUND

[0002] In the field of medical imaging, photon counting computed tomography (CT) imaging technology is an important detection means. This technology emits X-rays to a target object and measures the intensity of the X-rays after penetrating the object, and then uses a computer algorithm to reconstruct the structure image inside the object. However, in practical applications, CT imaging technology faces many challenges. For example, the inherent spectral characteristics of photon counting CT technology enable accurate identification of materials through material decomposition, but the noise of the base material reconstruction image is amplified during material decomposition, which directly affects the quality of virtual single-energy imaging, leading to a decrease in image clarity and contrast, thereby affecting the identification and diagnosis of lesions by doctors. In addition, due to the effects of pulse pile-up under high flux of the detector and the readout characteristics of the circuit, there is a large deviation between the real noise of the detector and the noise predicted using the traditional Poisson noise model. The inaccuracy of this noise model can lead to incorrect estimation and processing of noise during image reconstruction, further affecting the quality and reliability of the reconstructed image. SUMMARY

[0003] In view of the above-mentioned shortcomings of the prior art, the present application provides an image reconstruction method and system based on self-supervised denoising of a photon counting detector, a medium, a program product and a terminal, which are used to solve the problem of noise amplification during material decomposition in the prior art and the problem of poor accuracy of noise estimation using a traditional noise estimation model.

[0004] To achieve the above object and other related objects, the first aspect of the present application provides an image reconstruction method based on self-supervised denoising of a photon counting detector, comprising: acquiring energy spectrum CT projection data of a measured object based on a photon counting detector; performing material decomposition on the energy spectrum CT projection data to obtain a base material image; using a self-supervised denoising network combined with a denoising method based on deep learning noise prediction to denoise the base material image to obtain a denoised base material image; and reconstructing an energy spectrum CT image based on the denoised base material image using a reconstruction algorithm.

[0005] In some embodiments of the first aspect of the present application, the denoising method of the self-supervised denoising network combined with deep learning-based noise prediction is used to denoise the base material image to obtain a denoised base material image, and the process includes: inputting the base material image into the self-supervised denoising network to obtain a network estimation result of the base material image; and using a base material noise result predicted by a deep learning-based noise prediction method to optimize and denoise the network estimation result of the base material image to obtain a denoised base material image.

[0006] In some embodiments of the first aspect of the present application, the base material noise result predicted by the deep learning-based noise prediction method includes the following specific prediction process: collecting a plurality of sets of calibration data of base material combinations based on a photon counting detector; calculating the mean photon count of each pixel in the photon counting detector based on the plurality of sets of calibration data; inputting the mean photon count of each pixel into a detector count domain noise estimation network for prediction to obtain the variance of each pixel; and inputting the variance of each pixel into a noise propagation model to output a base material noise result.

[0007] In some embodiments of the first aspect of the present application, the detector count domain noise estimation network includes: a first convolutional layer, a fully connected layer combination unit, a second convolutional layer, and a third convolutional layer; wherein the first convolutional layer, the fully connected layer combination unit, the second convolutional layer, and the third convolutional layer are connected in sequence; the first convolutional layer, the second convolutional layer, and the third convolutional layer are each connected with a ReLU activation function; and the fully connected layer combination unit includes a plurality of fully connected layers, each of which is connected with a Tanh activation function.

[0008] In some embodiments of the first aspect of the present application, the calculation formula of the noise propagation model is: wherein l is the current base material, μ is the linear attenuation coefficient of the base material, M is the decomposition coefficient matrix of the base material in the energy interval, k is the covariance of the base material, and N is the attenuation value of the base material.

[0009] In some embodiments of the first aspect of the present application, the base material combination is: combining high-density material plates of different thicknesses and low-density material plates of different thicknesses in a predetermined combination manner.

[0010] To achieve the above object and other related objects, the second aspect of the present application provides a photon counting detector self-supervised denoising image reconstruction system, comprising: a data acquisition module configured to acquire energy spectrum CT projection data of an object to be measured based on a photon counting detector; a material decomposition module configured to perform material decomposition on the energy spectrum CT projection data to obtain a base material image; an image denoising module configured to perform denoising on the base material image using a self-supervised denoising network combined with a deep learning-based noise prediction denoising method to obtain a denoised base material image; and an image reconstruction module configured to reconstruct an energy spectrum CT image based on the denoised base material image using a reconstruction algorithm.

[0011] To achieve the above object and other related objects, the third aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the photon counting detector self-supervised denoising image reconstruction method.

[0012] To achieve the above object and other related objects, the fourth aspect of the present application provides a computer program product comprising computer program code, which, when executed on a computer, causes the computer to implement the photon counting detector self-supervised denoising image reconstruction method.

[0013] To achieve the above object and other related objects, the fifth aspect of the present application provides an electronic terminal comprising a memory, a processor, and a computer program stored on the memory; the processor executes the computer program to implement the photon counting detector self-supervised denoising image reconstruction method.

[0014] As described above, the photon counting detector self-supervised denoising image reconstruction method, system, medium, program product, and terminal provided by the present application have the following beneficial effects:

[0015] The present application constructs a noise propagation model of photon counting CT technology in the material decomposition process, improves the accuracy of detector count noise prediction based on neural networks, then optimizes the self-supervised denoising result using the noise propagation model, and solves the problem of insufficient training data of current photon counting CT. The present application does not require complex modeling processes and time-consuming calculation processes, but can accurately fit the response rules of different photon counting detectors with only a small amount of data, solve the noise amplification problem of PCCT technology in the material decomposition and reconstruction image process, and greatly improve the image quality and processing efficiency. The deep learning-based denoising method of the present application has shown very high practical value in the field of clinical application, and is expected to provide more reliable and high-quality technical support for medical diagnosis and other related work, and promote the further development of clinical practice. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of a method for image reconstruction based on self-supervised denoising of a photon counting detector according to an embodiment of the present application is shown.

[0017] Figure 2 A schematic diagram of the working principle of a conventional energy integrating detector and a photon counting detector according to an embodiment of the present application is shown.

[0018] Figure 3 A schematic diagram of a material decomposition result according to an embodiment of the present application is shown.

[0019] Figure 4 A schematic diagram of a structure for data acquisition of base material combinations according to an embodiment of the present application is shown.

[0020] Figure 5 A schematic diagram of a structure of a detector count domain noise estimation network according to an embodiment of the present application is shown.

[0021] Figure 6 A schematic diagram of the relative error distribution of a Poisson noise model and real noise according to an embodiment of the present application is shown.

[0022] Figure 7 A schematic diagram of the relative error distribution of the predicted values of two noise models and real noise according to an embodiment of the present application is shown.

[0023] Figure 8 A specific embodiment diagram of a method for image reconstruction based on self-supervised denoising of a photon counting detector according to an embodiment of the present application is shown.

[0024] Figure 9 A schematic diagram of a comparison result of reconstructed images of base material 1 and base material 2 at 200 mA tube current according to an embodiment of the present application is shown.

[0025] Figure 10 A schematic diagram of virtual mono-energy image reconstruction results of different methods at 60 keV according to an embodiment of the present application is shown.

[0026] Figure 11 A schematic diagram of a structure of an image reconstruction system based on self-supervised denoising of a photon counting detector according to an embodiment of the present application is shown.

[0027] Figure 12 A schematic diagram of a structure of an electronic terminal according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0028] The present application is illustrated by way of specific examples below, and other advantages and effects of the present application can be easily understood by those skilled in the art from the disclosure herein. The present application can also be implemented or applied in other different specific embodiments, and various modifications or changes can be made to the details herein based on different views and applications without departing from the spirit of the present application. It should be noted that the following examples and features in the examples can be combined with each other without conflict.

[0029] Before the present application is further described, the nomenclature and terminology used in the embodiments of the present application are explained, which are applicable to the following explanations:

[0030] <1> Photon counting detector: A photon counting detector (PCD) is a device used to detect X-ray photons, which can directly convert photon energy into an electrical signal, thereby realizing the counting and measurement of photons. Based on photon counting CT (PCCT) technology, additional spectral imaging information can be provided to improve imaging quality while reducing radiation dose. Compared with energy-integrating detectors (EIDs), PCDs have high energy conversion efficiency, good imaging quality, delicate structure design, and wide application range, and have broad application prospects in ultra-low dose CT, specific disease testing, and industrial detection.

[0031] <2> Photon counting CT technology: PCCT is a new type of medical imaging technology that uses photon counters to count X-rays to obtain higher spatial resolution and lower radiation dose. Compared with traditional CT, photon counting CT can obtain higher spatial resolution and lower radiation dose.

[0032] <3> Deep learning method: Deep learning method has very strong feature extraction and recognition ability, and has made great breakthroughs in various imaging application fields such as image denoising, deblurring, super-resolution, segmentation, classification, detection, recognition, etc. After systematic data training, it can spontaneously recognize and identify objects, and can replace the problems that need to be solved by artificial means in the past to some extent, improving the efficiency and accuracy of related research work.

[0033] <4>Neighbor2Neighbor algorithm: Neighbor2Neighbor algorithm is a self-supervised learning method for image denoising. It can train an effective denoising model only with noisy images, without relying on a large number of clean images as training data. The core idea of Neighbor2Neighbor algorithm is to use the similarity between local pixels in the image for self-supervised denoising. Specifically, it assumes that the values of adjacent pixels in the image have certain similarity, so it can use this similarity to construct a denoising model. In the training process, the model learns how to extract clean information from noisy images, so as to achieve the purpose of denoising.

[0034] <5>Filtered Back Projection (FBP): Filtered Back Projection is a reconstruction algorithm widely used in computed tomography (CT) and other imaging technologies. This algorithm is based on two key steps of back projection and filtering to reconstruct the internal structure of the object from a series of projection data.

[0035] To facilitate understanding of the embodiments of the present application, first, a kind of energy spectrum CT imaging method based on photon counting detector is introduced Figure 1 Detailed description. Figure 1 A flowchart of an image reconstruction method based on photon counting detector self-supervised denoising according to an embodiment of the present application is shown. The method in this embodiment includes:

[0036] Step S11: acquiring energy spectrum CT projection data of the measured object based on the photon counting detector.

[0037] It should be noted that, as Figure 2 shown, in the traditional energy integrating detector (EID), the incident X-ray photons are converted into clusters of visible light photons in the scintillator. The visible light irradiates the underlying photosensor, where positive and negative charges are generated. In the photon counting detector (PCD), X-ray photons are absorbed by a semiconductor material, generating positive and negative charges in the semiconductor material. Under the influence of a strong electric field, the positive and negative charges are pulled in opposite directions, generating an electrical signal.

[0038] In traditional energy-integrating detectors, the detector integrates the signal over a period of time. In photon-counting detectors, however, the detector directly measures the energy of each photon event. Compared to energy-integrating detectors, photon-counting detectors offer advantages such as high spatial resolution, low noise, low dose, and color imaging. Computed Tomography (CT) is a medical imaging method based on photon counting technology. This technology is primarily used to acquire high-resolution three-dimensional images of the human body's interior to aid doctors in diagnosis and treatment planning. This technology involves the photon-counting detector's sensitivity to X-rays, accurately recording the passage of each photon, which helps improve image quality and reduce radiation dose.

[0039] This invention utilizes photon-counting CT technology, employing a photon-counting detector to acquire projection data during X-ray energy spectrum computed tomography (ECT) imaging of the object under test. Specifically, the photon-counting detector receives and converts X-ray photons transmitted through the object under test in energy ranges. These photons are converted into electrical signals representing the attenuation information of X-rays interacting with the object at different energies, i.e., energy spectrum CT projection data. This energy spectrum CT projection data can be used for subsequent energy spectrum analysis and image reconstruction.

[0040] Step S12: Perform material decomposition on the energy spectrum CT projection data to obtain the base material image.

[0041] It's important to note that the information collected by photon counting detectors is typically a composite of different substances. For example, in the medical field, human tissue contains a variety of different components, such as bones, muscles, blood, fat, and potentially diseased tissue; in industrial inspection, the object being inspected may also be a complex structure composed of multiple materials. Through material decomposition, this complex composite information can be broken down into information about different base materials, helping to gain a deeper understanding of the fundamental materials that make up the object under test, thereby accurately identifying its composition.

[0042] Specifically, based on the different absorption characteristics of X-rays by different materials, the absorption information of different materials can be separated through material decomposition methods, decomposing them into different base material images. This can more clearly display the distribution of different base materials, enhance image contrast and visualization effects, and clearly show information that was originally mixed and difficult to distinguish in different base material images, making it easier for users to observe and analyze. At the same time, the complex multi-material imaging problem can be transformed into multiple single-material imaging problems. Analyzing and processing single-material images can simplify the image reconstruction process and improve reconstruction accuracy and reconstructed image quality.

[0043] Step S13: denoising the base material image by using a self-supervised denoising network combined with a deep learning-based noise prediction denoising method to obtain a denoised base material image.

[0044] In an embodiment, the process of denoising the base material image by using a self-supervised denoising network combined with a deep learning-based noise prediction denoising method to obtain a denoised base material image comprises: inputting the base material image into the self-supervised denoising network to obtain a network estimation result of the base material image; and optimizing and denoising the network estimation result of the base material image by using a base material noise result predicted by the deep learning-based noise prediction method to obtain a denoised base material image.

[0045] For example, as shown in FIG. 5, the spectral CT projection data of the object to be measured obtained by scanning at a tube current of 50 mA is amplified in the material decomposition process, and the noise of the base material image is enlarged. Figure 3 As shown in FIG. 5, the reconstructed images of the base material 1 and the base material 2 obtained after material decomposition have obvious noise and poor image definition, thereby affecting the quality of subsequent virtual mono-energy imaging. Therefore, in the present embodiment, a denoising method for the base material image is proposed, and the specific denoising process is to obtain an initial denoising result of the base material image, i.e., a network estimation result, according to a self-supervised denoising network, and then optimize the initial denoising result of the base material image by using prior information (a base material noise result) predicted by a deep learning-based noise prediction method, thereby improving the denoising effect and obtaining a final denoised base material image.

[0046] Preferably, the self-supervised denoising network uses a Neighbor2Neighbor algorithm. In a traditional image denoising algorithm, a training model using a supervised denoising algorithm relies on clean and noise-free images to optimize image denoising, and performs well under the use of a large amount of clean image data. However, in a real scene, it is difficult to obtain clean and noise-free images. The Neighbor2Neighbor algorithm does not require clean and noise-free images, and it trains a pair of images by generating images, i.e., a pair of neighborhood images of a single noisy image, and therefore has the advantages of high training efficiency, small data volume and excellent performance.

[0047] In the present embodiment, the base material image is input into the self-supervised denoising network to obtain a network estimation result of the base material image, the network estimation result includes a noise estimation result of the base material image, and a deep learning-based noise prediction method is used to predict a base material noise result. The noise estimation result of the base material image is optimized and updated by using a maximum log-likelihood method, so as to be closer to the base material noise result, thereby realizing denoising of the base material image and obtaining a denoised base material image.

[0048] In an embodiment, the base material noise result predicted by the deep learning-based noise prediction method comprises the following specific prediction process: a plurality of sets of calibration data of the base material combination are collected based on a photon counting detector; the mean photon count of each pixel in the photon counting detector is calculated according to the plurality of sets of calibration data; the mean photon count of each pixel is input into a detector count domain noise estimation network for prediction to obtain the variance of each pixel; and the variance of each pixel is input into a noise propagation model to output the base material noise result.

[0049] In combination Figure 4 It is explained that the acquisition system structure for collecting a plurality of sets of calibration data of the base material combination is a ray source, a base material combination and a photon counting detector. The ray source emits X-ray photons to irradiate the base material combination, and the detector collects the X-ray photons after transmitting through the base material combination. The collected photon data is called calibration data. The base material combination is a combination of high-density material plates with different thicknesses and low-density material plates with different thicknesses in a preset combination manner. For example, the preset combination manner is to select a 20mm high-density material plate and a 120mm low-density material plate for combination, or to select a 30mm high-density material plate and a 100mm low-density material plate for combination. According to the actual situation, the final acquisition of a plurality of base material combinations is not limited in this embodiment.

[0050] Specifically, a high-density material plate (such as aluminum, copper, lead, bismuth, tungsten, etc.) with a fixed thickness is combined with a low-density material plate (such as aluminum oxide, polyethylene, polypropylene, styrene foam, etc.) with a fixed thickness to form a base material combination. The base material combination is irradiated by a ray source to obtain the calibration data of the detector. Then, the thicknesses of the high-density material plate and the low-density material plate are changed to recombine the base material combination. The above acquisition process is repeated to obtain a plurality of sets of calibration data.

[0051] Further, the plurality of sets of calibration data currently obtained are calculated to obtain the mean photon count of each pixel in the photon counting detector. It should be understood that the photon counting detector includes a plurality of pixels. When the base material combination is scanned multiple times, the photon counts of each pixel corresponding to the photon counting detector are different, and the mean photon count is calculated. At the same time, it is also necessary to consider that the emission spectrum of the ray source is divided into different energy intervals, that is, the mean photon count of each pixel in the photon counting detector for each energy interval is calculated. The mean photon count reflects the average level of the number of photons received by the detector within a certain time or region, and is used for subsequent prediction of the variance of each pixel.

[0052] The mean photon count is then input into the detector counting domain noise estimation network for prediction, thereby predicting the variance of each pixel. The detector counting domain noise estimation network used in this invention is a deep learning-based neural network architecture. It obtains the noise distribution of each energy channel from the photon counting detector data, and the neural network learns the characteristics of this distribution. This network is designed to estimate the noise in the detector counting domain, i.e., predict the variance of the photon count collected for each pixel. Leveraging the powerful feature extraction and learning capabilities of deep learning neural networks, the detector counting domain noise estimation network can more accurately predict the noise variance of each pixel, achieving higher estimation accuracy compared to traditional noise estimation methods.

[0053] In some examples, such as Figure 5 As shown, the detector counting domain noise estimation network is a neural network structure, including: a first convolutional layer, a fully connected layer combination unit, a second convolutional layer, and a third convolutional layer; wherein, the first convolutional layer, the fully connected layer combination unit, the second convolutional layer, and the third convolutional layer are connected sequentially; a ReLU activation function is connected after the first convolutional layer, the second convolutional layer, and the third convolutional layer; the fully connected layer combination unit includes multiple fully connected layers; and a Tanh activation function is connected after each fully connected layer.

[0054] By inputting the predicted variance into the noise propagation model, the noise results for the base material can be obtained. Specifically, the noise in the detector energy channel is converted into the noise of the base material decomposition coefficients, thereby deriving the noise propagation model from the detector counting domain to the material decomposition projection domain. The derivation process is as follows:

[0055] First, it should be understood that photon counting CT technology is an imaging technique that uses the attenuation differences of X-rays after penetrating different materials. The process of calculating the attenuation coefficient distribution from the projection data is image reconstruction. The pixel values ​​of the reconstructed image represent the attenuation coefficient of the X-rays, ultimately revealing the internal structure and composition of the object being measured. The linear attenuation coefficient of X-ray materials is mainly composed of Compton scattering and the photoelectric effect. The formula for calculating the linear attenuation coefficient of X-ray materials is as follows:

[0056]

[0057] in, The linear decay coefficient is at energy E and position vector is The distribution of α is the decomposition coefficient, and μ is the linear decay coefficient.

[0058] When two different substances are used as base materials The formula can be further expressed as:

[0059]

[0060] Wherein, l i is the effective path length of the material, i is the base material type.

[0061] In the actual scanning scene of the measured object, the material decomposition form of the PCCT spectrum response model is as follows:

[0062]

[0063] Wherein, N 0i is the unattenuated value, Omega (E) is the normalized spectrum, N i is the attenuated value. Further, the calculation formula of the projection value corresponding to the energy is as follows:

[0064]

[0065] Since p i is continuously differentiable, its Taylor expansion at (k1, k2) is as follows:

[0066]

[0067] The present application studies the correlation between and , which is expressed by a first-order approximation noise term as follows:

[0068]

[0069] The covariance matrix formula of the base material is as follows:

[0070]

[0071] Further, in the prior art, the noise estimation is usually performed by using the method of Poisson model, but due to the pulse pile-up effect under the high flux condition of the photon counting detector and the influence of the readout characteristics of the circuit, the traditional Poisson noise model is insufficient, and there is a large deviation from the real noise model of the photon counting detector. As shown in Figure 6 , the average error between the Poisson model and the real noise reaches about 10%, and the maximum error exceeds 20%, and the noise estimation effect is deviated. According to the above formula, the noise model based on the Poisson model can be derived as follows:

[0072]

[0073] And in the present embodiment, the noise propagation model based on deep learning is used, and the calculation formula of the derived noise propagation model is as follows:

[0074]

[0075] Where l is the current base material, μ is the linear decay coefficient of the base material, M is the decomposition coefficient matrix of the base material in the energy range, k is the covariance of the base material, and N is the decay value of the base material.

[0076] For example, this invention compares the noise estimation results of a noise model based on the Poisson model and a noise propagation model based on deep learning. The comparison results are as follows: Figure 7 As shown, where, Figure 7 (a) shows the relative error distribution between the predicted values ​​of the noise model based on the Poisson model and the noise propagation model based on deep learning and the actual noise values ​​in the energy range 1 (30-60keV). Figure 7 (b) presents the relative error distributions between the predicted values ​​and actual noise values ​​of the Poisson-based noise model and the deep learning-based noise propagation model, respectively, within the energy range of 2 (60-120 keV). Figure 7 It is clear that the predictions of the noise propagation model based on deep learning are closer to the actual noise, meaning the error between them is smaller. Therefore, the noise propagation model based on deep learning significantly outperforms the noise model based on the Poisson model in noise estimation.

[0077] It should be understood that the self-supervised denoising network proposed in this invention, combined with a deep learning-based noise prediction denoising method, first constructs a deep learning-based neural network architecture to estimate the true noise of each pixel in the detector based on the received photon count. Then, it constructs a calculation formula for a noise propagation model to derive the base material noise. This formula describes the influence of photon count fluctuations on material noise during material decomposition, thus converting the predicted detector noise into base material noise. Finally, the detector noise propagation model is integrated into the self-supervised denoising network, and an optimization strategy is constructed using the log-likelihood maximization method for denoising the actual measurement data of the object under test. This denoising method can improve the accuracy of PCCT material decomposition and the quality of virtual monochrome imaging, while requiring only a small amount of experimental data.

[0078] This application eliminates the need for complex modeling processes and time-consuming calculations. It accurately fits the response patterns of different photon counting detectors using only a small amount of data, solving the noise amplification problem in PCCT technology during material decomposition and image reconstruction, thus significantly improving image quality and processing efficiency. The deep learning-based denoising method of this invention demonstrates extremely high practical value in clinical applications and is expected to provide more reliable and high-quality technical support for medical diagnosis and related work, promoting further development in clinical practice.

[0079] Step S14: Reconstruct the energy spectrum CT image based on the noise-reduced base material image using a reconstruction algorithm.

[0080] It should be noted that the reconstruction algorithm is preferably a filtered back projection algorithm. The filtered back projection algorithm is a three-dimensional imaging algorithm controlled by a computer system in an iterative manner. Compared with the traditional CT scanning method, the FBP algorithm adopts an iterative manner for reconstruction, avoids multiple projections, thereby reducing the scanning time; the FBP algorithm can effectively suppress noise; the image reconstructed by the FBP algorithm has good quality and is not sensitive to noise.

[0081] The denoised base material image significantly reduces noise interference and has high image quality. The denoised base material image is processed by using an image reconstruction algorithm, reconstruction from the denoised base material image to the spectral CT image is realized, and the obtained spectral CT image can accurately reflect the spectral information and internal structural characteristics of the measured object, thereby providing high-quality image data support for subsequent medical diagnosis, material analysis and other applications.

[0082] In order to facilitate the display of the image reconstruction method based on the photon counting detector self-supervised denoising provided in the present application, the following specific embodiments are provided for illustration in combination with Figure 8 , Figure 9 and Figure 10 .

[0083] The photon counting detector is composed of a 912x16 pixel array, and spectral CT projection data of a head phantom and a GAMMEX phantom are collected at a tube current of 50mA and 200mA. The emission spectrum of the ray source includes two energy intervals: 30-60keV and 60-120keV. The head phantom and the GAMMEX phantom are both composed of two base materials: PMMA (acrylic) and Al (aluminum).

[0084] As shown in Figure 8 , the image reconstruction method based on the photon counting detector self-supervised denoising has the following process: the acquired spectral CT projection data is subjected to material decomposition respectively to obtain a base material 1 (PMMA) image and a base material 2 (Al) image, the base material images are input into a self-supervised denoising network to obtain a network estimation result (μ,Σ μ ), i.e. the initial denoising result of the base material image; a plurality of sets of calibration data of the base material combination are obtained, the calibration data is calculated to obtain the photon counting mean value α(n1,n2) of each pixel, the photon counting mean value is input into a detector counting domain noise estimation network to obtain the predicted variance Σ(α1,α2), the predicted variance is input into a noise propagation model to obtain the base material noise result Σ n (l1,l2); the network estimation result (μ,Σ n (l1,l2) is corrected based on the base material noise result Σ μ) an optimized update is performed, that is, an initial denoising result of the base material image is optimized to obtain a final denoised base material image; and the FBP is used to reconstruct the denoised base material image to obtain a reconstructed image.

[0085] As shown in Figure 9 , a comparison result of the reconstructed images of the base material 1 (PMMA) and the base material 2 (Al) at a tube current of 200 mA is shown, and the display range is [0, 1500] mg / mL for PMMA and [0, 16] mg / mL for Al. It can be seen that, in the image reconstruction of the head phantom, if the FBP method is used for the image reconstruction of the base material 1 and the base material 2, there is still a large amount of noise in the reconstructed image, which affects the image quality and the visual effect of the image is poor and unclear; if the base material image denoised by the Proposed-P (noise model based on Poisson model) denoising method is reconstructed based on the FBP method, the reconstructed image is obviously improved compared with the image reconstructed by the FBP method alone, but compared with the reconstructed image denoised by the Proposed-L (noise propagation model based on deep learning) denoising method of the present application, the image of the present application is obviously cleaner and clearer; in the comparison of the image reconstruction results of the GAMMEX phantom, the conclusion is consistent with the above-mentioned head phantom image reconstruction conclusion.

[0086] As shown in Figure 10 , the virtual mono-energy image reconstruction results of different tube currents (50 mA and 200 mA) and different methods (FBP, Proposed-P and Proposed-L) at an energy of 60 keV are shown, and the display range of the GAMMEX phantom is [-100, 100] HU and the display range of the head phantom is [-60, 60] HU. It can be seen that, for the GAMMEX phantom and the head phantom, the best denoising method is still Proposed-L whether the tube current is 200 mA or 50 mA, and the visual effect of the image reconstructed after denoising by Proposed-L is clearer and has less noise.

[0087] From the above, it can be seen that the denoising method of the present application has better denoising effect and higher reconstructed image quality.

[0088] In the embodiments of the present application, the same items or similar items with basically the same functions and effects are distinguished by using "first", "second" and the like, and the sequence is not limited. Those skilled in the art can understand that "first", "second" and the like do not limit the number and execution sequence, and "first", "second" and the like do not necessarily mean different.

[0089] It should be noted that in the embodiments of the present application, the words "exemplary" or "for example" mean an example, an illustration or an illustration. Any embodiment or design scheme described as "exemplary" or "for example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the use of "exemplary" or "for example" and the like is intended to present the relevant concept in a specific manner.

[0090] In the embodiments of the present application, "at least one" means one or more, and "multiple" means two or more. The association relationship between the associated objects is described, which means that there can be three kinds of relationships, for example, A and / or B, which can represent the following cases: A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after it. "At least one of the following" or similar expressions means any combination of these items, including any combination of single item or multiple items. For example, at least one of a, b or c can represent a, b, c, a-b, a-c, b-c or a-b-c, where a, b and c can be single or multiple.

[0091] Figure 11 is a schematic block diagram of an image reconstruction system based on photon counting detector self-supervised denoising provided by the embodiments of the present application. As shown in Figure 11 , the system 1100 includes:

[0092] The data acquisition module 1101 is configured to acquire energy spectrum CT projection data of a measured object based on a photon counting detector.

[0093] The material decomposition module 1102 is configured to perform material decomposition on the energy spectrum CT projection data to obtain a base material image.

[0094] The image denoising module 1103 is configured to denoise the base material image using a self-supervised denoising network combined with a noise prediction based on deep learning to obtain a denoised base material image.

[0095] The image reconstruction module 1104 is configured to reconstruct an energy spectrum CT image based on the denoised base material image using a reconstruction algorithm.

[0096] It should be understood that the specific processes of each module for performing the corresponding steps described above have been described in detail in the above method embodiments, and for the sake of brevity, will not be repeated here.

[0097] It should also be appreciated that the division of tasks between the modules in the embodiments of the present application is illustrative only and that other divisions of tasks between the modules are possible. In addition, the various functional modules in the various embodiments of the present application can be integrated in one processor, or can exist separately, or two or more modules can be integrated in one module. The integrated modules can be implemented in the form of hardware or in the form of software functional modules.

[0098] Figure 12 is a schematic block diagram of an electronic terminal provided by an embodiment of the present application. As shown in Figure 12 the electronic terminal includes at least one processor 1201, a memory 1202, at least one network interface 1203 and a user interface 1205. The various components in the apparatus are coupled together by a bus system 1204. It can be understood that the bus system 1204 is used to realize the connection and communication between the components. The bus system 1204 includes a data bus in addition to power buses, control buses and status signal buses. However, in order to clearly illustrate, all the buses are marked as the bus system in Figure 12 .

[0099] The user interface 1205 can include a display, a keyboard, a mouse, a trackball, a click gun, a key, a button, a touchpad or a touch screen, etc.

[0100] It can be understood that the memory 1202 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. The non-volatile memory can be a read-only memory (ROM, Read Only Memory), a programmable read-only memory (PROM, Programmable Read-Only Memory), which is used as an external cache. By way of example but not limitation, many forms of RAM can be used, such as static random access memory (SRAM, Static Random Access Memory), synchronous static random access memory (SSRAM, Synchronous Static Random Access Memory). The memory described in the embodiments of the present application is intended to include but not limited to these and any other suitable categories of memory.

[0101] The memory 1202 in the embodiments of the present application is configured to store various types of data to support the operation of the electronic terminal 1200. Examples of the data include any executable programs for operating on the electronic terminal 1200, such as an operating system 12021 and an application program 12022. The operating system 12021 includes various system programs, such as a framework layer, a core library layer, a driver layer, and the like, for implementing various basic services and processing hardware-based tasks. The application program 12022 can include various application programs, such as a media player, a browser, and the like, for implementing various application services. The application program 12022 can include the method for implementing the image reconstruction based on the photon counting detector self-supervised denoising provided by the embodiments of the present application.

[0102] The method disclosed in the embodiments of the present application can be applied to the processor 1201 or implemented by the processor 1201. The processor 1201 can be an integrated circuit chip having a processing capability of signals. In the implementation process, each step of the above method can be completed by the integrated logic circuit of hardware in the processor 1201 or the instruction in the form of software. The processor 1201 described above can be a general processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and the like. The processor 1201 can implement or execute the disclosed methods, steps, and logic block diagrams in the embodiments of the present application. The general processor 1201 can be a microprocessor or any conventional processor, and the like. In combination with the steps of the accessory optimization method provided by the embodiments of the present application, the hardware decoding processor can be directly embodied to complete the execution, or the combination of hardware and software modules in the decoding processor can be used to complete the execution. The software module can be located in a storage medium, which is located in the memory. The processor reads the information in the memory and combines the hardware to complete the steps of the above method.

[0103] In the exemplary embodiments, the electronic terminal 1200 can be one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), and the like, for executing the above method.

[0104] According to the method provided in the embodiments of the present application, the present application further provides a computer program product, which comprises computer program codes, and when the computer program codes run on a computer, the computer is caused to execute the image reconstruction method based on the photon counting detector self-supervised denoising according to any one of the embodiments.

[0105] According to the method provided in the embodiments of the present application, the present application further provides a computer readable storage medium, which stores program codes, and when the program codes run on a computer, the computer is caused to execute the image reconstruction method based on the photon counting detector self-supervised denoising according to any one of the embodiments.

[0106] The terms "component", "module", "system", and the like used in the present specification are used to represent computer-related entities, hardware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable, a thread of execution, a program, and / or a computer. By way of illustration, both an application running on a computing device and the computing device can be a component. One or more components can reside within a process and / or thread of execution, and a component can be localized on one computer and / or distributed between two or more computers. In addition, these components can execute from various computer readable media having various data structures stored thereon. The components can communicate by way of local and / or remote processes such as in accordance with a signal having one or more data packets (e.g., data from one component interacting with another component in a local system, distributed system, and / or across a network such as the Internet with other systems via the signal).

[0107] Those of ordinary skill in the art can realize that the various illustrative logical blocks and steps described in connection with the embodiments disclosed herein can be implemented or performed by electronic hardware, or a combination of computer software and electronic hardware. The functions described herein can be performed by hardware, software, or a combination thereof, depending on the particular application and design constraints. Those of ordinary skill in the art can realize the mechanisms described herein with a different method, or other functions, without departing from the scope of the present application.

[0108] Those of ordinary skill in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the system, device and unit described above can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0109] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the described device embodiments are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units can be indirect couplings or communication connections through some interfaces, devices or units, and can be electrical, mechanical or in other forms.

[0110] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they can be located in one place, or distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0111] In addition, the functional units in each embodiment of the present application can be integrated into one processing unit, or each unit can be a physically separate unit, or two or more units can be integrated into one unit.

[0112] In the above embodiments, the functions of the functional units can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, the functions can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions (programs). When the computer program instructions (programs) are loaded and executed on a computer, the whole or part of the processes or functions according to the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transferred from one website site, computer, server or data center to another website site, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be magnetic media (such as floppy disk, hard disk, magnetic tape), optical media (such as high-density digital video disc (digital video disc, DVD), or semiconductor media (such as solid state disk (solid state disk, SSD), etc.

[0113] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0114] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0115] In summary, the present application provides a photon counting detector self-supervised denoising image reconstruction method, system, medium, program product and terminal. By constructing a noise propagation model of photon counting CT technology in the material decomposition process, the accuracy of detector count noise prediction is improved based on a neural network, then the noise propagation model is used to optimize the self-supervised denoising result, and the problem of insufficient training data of current photon counting CT is solved. The present application does not need complex modeling process and time-consuming calculation process, and can accurately fit the response law of different photon counting detectors with a small amount of data, solves the noise amplification problem of PCCT technology in the material decomposition and reconstruction image process, greatly improves the image quality and processing efficiency. The deep learning based denoising method of the present application has shown very high practical value in the field of clinical application, and is expected to provide more reliable and high-quality technical support for medical diagnosis and other related work, and promote the further development of clinical practice. Therefore, the present application effectively overcomes the various shortcomings in the prior art and has high industrial utilization value.

[0116] The above embodiments only exemplarily illustrate the principles and effects of the present application, and are not used to limit the present application. Any person skilled in the art can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by those skilled in the art without departing from the spirit and technical idea disclosed in the present application should be covered by the claims of the present application.

Claims

1. An image reconstruction method based on self-supervised denoising using a photon counting detector, characterized in that, include: Energy spectrum CT projection data of the object under test are acquired based on a photon counting detector; Material decomposition is performed on the energy spectrum CT projection data to obtain an image of the base material; A self-supervised denoising network combined with a deep learning-based noise prediction denoising method is used to denoise the base material image to obtain a denoised base material image. Energy spectrum CT images are reconstructed based on the noise-reduced base material images using a reconstruction algorithm.

2. The image reconstruction method based on self-supervised denoising using a photon counting detector according to claim 1, characterized in that, The process of denoising the base material image using a self-supervised denoising network combined with a deep learning-based noise prediction method to obtain the denoised base material image includes: The base material image is input into a self-supervised denoising network to obtain the network estimation result of the base material image; The noise prediction results of the base material image are optimized and denoised using the noise prediction results of the base material based on a deep learning-based noise prediction method to obtain the denoised base material image.

3. The image reconstruction method based on self-supervised denoising using a photon counting detector according to claim 2, characterized in that, The noise prediction results of the base material predicted by the noise prediction method based on deep learning include the following specific prediction process: Multiple sets of calibration data were collected based on a photon counting detector and a combination of base materials. The average photon count of each pixel in the photon counting detector is calculated based on the multiple sets of calibration data. The mean photon count of each pixel is input into the detector counting domain noise estimation network for prediction to obtain the variance of each pixel; The variance of each pixel is input into the noise propagation model to output the base material noise result.

4. The image reconstruction method based on self-supervised denoising using a photon counting detector according to claim 3, characterized in that, The detector counting domain noise estimation network includes: a first convolutional layer, a fully connected layer combination unit, a second convolutional layer, and a third convolutional layer; The first convolutional layer, the fully connected layer combination unit, the second convolutional layer, and the third convolutional layer are connected sequentially; each of the first, second, and third convolutional layers is followed by a ReLU activation function; the fully connected layer combination unit includes multiple fully connected layers; and each fully connected layer is followed by a Tanh activation function.

5. The image reconstruction method based on self-supervised denoising using a photon counting detector according to claim 3, characterized in that, The calculation formula for the noise propagation model is as follows: Where l is the current base material, μ is the linear decay coefficient of the base material, M is the decomposition coefficient matrix of the base material in the energy range, k is the covariance of the base material, and N is the decay value of the base material.

6. The image reconstruction method based on self-supervised denoising using a photon counting detector according to claim 3, characterized in that, The base material combination is: combining high-density material plates of different thicknesses and low-density material plates of different thicknesses according to a preset combination method.

7. An image reconstruction system based on self-supervised denoising using a photon counting detector, characterized in that, include: The data acquisition module is used to acquire energy spectrum CT projection data of the object under test based on a photon counting detector; The material decomposition module is used to decompose the energy spectrum CT projection data into a base material image. The image denoising module is used to denoise the base material image by employing a self-supervised denoising network combined with a denoising method based on deep learning noise prediction, so as to obtain a denoised base material image. The image reconstruction module is used to reconstruct energy spectrum CT images based on the denoised base material images using a reconstruction algorithm.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the image reconstruction method based on self-supervised denoising of a photon counting detector as described in any one of claims 1 to 6.

9. A computer program product, characterized in that, The computer program product includes computer program code, which, when run on a computer, enables the computer to implement the image reconstruction method based on self-supervised denoising of a photon counting detector as described in any one of claims 1 to 6.

10. An electronic terminal, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the image reconstruction method based on self-supervised denoising of a photon counting detector as described in any one of claims 1 to 6.

Images