CN116011327A - Deep learning substance decomposition method, device, terminal and storage medium based on physical model - Google Patents
Deep learning substance decomposition method, device, terminal and storage medium based on physical model Download PDFInfo
- Publication number
- CN116011327A CN116011327A CN202211711964.8A CN202211711964A CN116011327A CN 116011327 A CN116011327 A CN 116011327A CN 202211711964 A CN202211711964 A CN 202211711964A CN 116011327 A CN116011327 A CN 116011327A
- Authority
- CN
- China
- Prior art keywords
- detector
- neural network
- response
- physical
- physical model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Measurement Of Radiation (AREA)
Abstract
The application provides a deep learning substance decomposition method based on a physical model, which comprises the following steps: building a physical model, which is used for mapping the responses of the X-rays of the CT detector under different physical parameters and manufacturing a data set based on the physical model; constructing and training a neural network based on the data set, and fitting the responses of the CT detector under different physical parameters; setting a plurality of calibration experiments, and determining physical parameters of the CT detector by reducing errors of a neural network predicted value and an experiment response value; and in the imaging process, calculating thickness information of the detected substance through a neural network according to physical parameters of the CT detector and actual response of the CT detector. According to the method, the neural network is built to map the correlation among the physical parameters in the material decomposition process of the CT detector, and the physical parameters of the CT detector are obtained through a small amount of calibration experiments, so that the thickness information of the detected material is reversely deduced through the known physical parameters and the response energy spectrum when the material is decomposed, and the method has good robustness.
Description
Technical Field
The present disclosure relates to the field of CT imaging, and in particular, to a method, an apparatus, a terminal, and a storage medium for decomposing a deep learning substance based on a physical model.
Background
Computer tomography (Computed Tomography, CT for short) is an indispensable examination means in modern medicine and plays an irreplaceable role in the diagnosis of many diseases. With the development of photon counting detector technology, photon counting energy spectrum CT has gained increasing attention. The energy spectrum CT refers to a CT imaging mode using X-ray energy spectrum information, and aims to eliminate the influence of energy in an imaging result by using the difference of material attenuation coefficients under different energy spectrums, so that the energy spectrum CT can provide material resolving power which is not available in the traditional CT and has important significance for contrast imaging and soft tissue imaging; and simultaneously, the noise is lower than that of dual-energy CT, so that the dosage of a patient can be reduced.
The process of substance identification using different X-ray energy spectrum information is called substance decomposition, and the core of the substance decomposition is to obtain substance thickness information of a detected object, so that imaging is performed based on the substance thickness information to restore internal structural information of the detected object. However, the existing photon counting detectors still have some defects which are hard to overcome physically, and this brings hidden danger to the imaging quality of photon counting CT. For example, signal stacking can result in increased image noise and destruction of the energy spectrum; compton scattering can cause errors in the detector energy count, resulting in errors in the final energy spectrum; the charge sharing effect can cause two adjacent detector units to repeatedly count the same photon, and errors can also be generated; the escape and reabsorption phenomena of characteristic X-rays can also lead to certain errors in the energy spectrum. The physical phenomena have different influences on the quality of substance decomposition, and are the problems to be solved in the prior art.
The existing CT substance decomposition methods mainly comprise the following three types:
1. and (5) an image method. The principle is that the traditional image reconstruction algorithm is used for obtaining images under different photon energies and then carrying out substance decomposition to obtain thickness information of detected substances, and the method can not directly obtain a substance decomposition result and has radiation hardening artifacts and amplification noise;
2. and (5) modeling. The principle is that the thickness information of the detected substance is directly obtained according to the count value of the detector by establishing a physical model, and the method has the defect that the parameters of the physical model are too complex, so that a large amount of calculation is needed, and real-time imaging cannot be performed;
3. calibration method. The principle is that the correlation between the counting response of the detector and the thickness information of the detected substance is obtained through a large amount of calibration experiments and calibration data, and the disadvantage is that the data calibration process is too complex and the efficiency is not high.
Disclosure of Invention
In view of the above-mentioned drawbacks of the prior art, an object of the present application is to provide a method, an apparatus, a terminal and a storage medium for deep learning substance decomposition based on a physical model, which are used for solving the technical problems of complex calculation and complex calibration of the existing substance decomposition method.
To achieve the above and other related objects, a first aspect of the present application provides a method comprising: building a physical model, which is used for mapping the responses of the X-rays of the CT detector under different physical parameters and manufacturing a data set based on the physical model; constructing and training a neural network based on the data set for fitting the response of the detector under different physical parameters; setting a plurality of calibration experiments, and determining physical parameters of the CT detector by reducing errors of a neural network predicted value and an experiment response value; and in the imaging process, calculating thickness information of the detected substance through a neural network according to physical parameters of the CT detector and actual response of the CT detector.
In some embodiments of the first aspect of the present application, the physical parameters of the CT detector are derived based on the following calibration experiments: acquiring a predicted value of a physical parameter of the CT detector, thickness information of an experimental substance and an experimental response spectrum of the CT detector; inputting the thickness information of the experimental material and the predicted value of the physical parameter of the CT detector into the neural network after the pre-training to obtain a predicted response spectrum of the CT detector; calculating an error between a predicted response spectrum of the CT detector and an experimental response spectrum of the CT detector, if the error is greater than a preset error threshold, updating a predicted value of a physical parameter of the CT detector, and repeating the steps until the error is less than the preset error threshold; and when the error is smaller than a preset error threshold, enabling the physical parameter of the CT detector to be equal to the predicted value of the physical parameter of the current CT detector.
In some embodiments of the first aspect of the present application, the pre-trained neural network is obtained based on the following steps: building a physical model, wherein the physical model is used for mapping a functional relation among an initial energy spectrum, physical parameters of a CT detector, thickness information of detected substances and a response energy spectrum of the CT detector; generating a data set according to the physical model, wherein the data set comprises physical parameters of different CT detectors under a fixed initial energy spectrum, thickness information of different detected substances and response energy spectrums of the CT detectors corresponding to the physical parameters and the thickness information; the data set is input to an untrained neural network for training the untrained neural network to converge.
In some embodiments of the first aspect of the present application, the physical parameters of the CT detector include detector thickness, detection efficiency, electric field strength, charge sharing coefficient, chip dead time, chip count threshold, chip threshold drift coefficient with count rate, energy resolution.
In some embodiments of the first aspect of the present application, the pre-trained neural network comprises a pre-trained first neural network and a pre-trained second neural network; the pre-trained first neural network is used for mapping a functional relation among detector thickness, detection efficiency, electric field intensity, charge sharing coefficient, thickness information of detected substances and an intermediate energy spectrum of the CT detector, and the pre-trained second neural network is used for mapping a functional relation among chip dead time, chip counting threshold, chip threshold drift coefficient with counting rate, energy resolution, thickness information of detected substances and a final response energy spectrum of the CT detector.
In some embodiments of the first aspect of the present application, the physical model includes a first physical model and a second physical model; the first physical model is used for mapping a functional relation among an initial energy spectrum, detector thickness, detection efficiency, electric field intensity, charge sharing coefficient, thickness information of detected substances and an intermediate energy spectrum of the CT detector, and the second physical model is used for mapping a functional relation among the initial energy spectrum, chip dead time, chip counting threshold, chip threshold drift coefficient with counting rate, energy resolution, thickness information of detected substances and a final response energy spectrum of the CT detector.
In some embodiments of the first aspect of the present application, the untrained neural network comprises an untrained first neural network and an untrained second neural network; when the untrained neural network is trained, the untrained first neural network is input into an ideal energy spectrum, the thickness of the detector, the detection efficiency, the electric field strength and the charge sharing coefficient, and output into a response energy spectrum of the detector under the thickness of the detector, the detection efficiency, the electric field strength and the charge sharing coefficient; the untrained second neural network input is the output of the untrained first neural network, the chip dead time, the chip count threshold, the chip threshold drift coefficient with count rate, the energy resolution, and the output is the actual response spectrum of the chip.
To achieve the above and other related objects, a second aspect of the present application provides a deep learning substance decomposition device based on a physical model, comprising: parameter acquisition module: the method comprises the steps of acquiring a response energy spectrum of a CT detector and physical parameters of the CT detector; the physical parameters of the CT detector are obtained by reducing the error between an experimental response spectrum of the CT detector and a predicted response spectrum of the CT detector through a preset calibration experiment; and a thickness calculation module: according to the response energy spectrum of the CT detector and the physical parameters of the CT detector, calculating thickness information of detected substances based on a neural network trained in advance; the pre-trained neural network is used for mapping the functional relation among the physical parameters of the CT detector, the thickness information of the detected substance and the response energy spectrum of the CT detector.
To achieve the above and other related objects, a third aspect of the present application provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method.
To achieve the above and other related objects, a fourth aspect of the present application provides an electronic terminal, including: a processor and a memory; the memory is used for storing a computer program, and the processor is used for executing the computer program stored in the memory so as to enable the terminal to execute the method.
As described above, the deep learning substance decomposition method, device, terminal and storage medium based on the physical model have the following beneficial effects:
according to the method, the neural network is built to map complex correlations among physical parameters in the photon counting CT detector substance decomposition process, the physical parameters of the CT detector are obtained through a small amount of calibration experiments, substance decomposition is performed on the basis of the known physical parameters and the trained neural network in practical application, and the most-conforming substance thickness information can be obtained rapidly. Compared with the traditional substance decomposition method, the method has the advantages that the calculation resources required to be equipped are obviously reduced, the operation efficiency is greatly improved, and the method is convenient for flexible deployment in practical application. It is emphasized that the method does not need to adopt complex calibration procedures in the traditional substance decomposition method, and can obtain ideal physical parameters of the CT detector only through a plurality of groups of calibration experiments, so that the overall efficiency of substance decomposition is greatly improved, the method has unusual accuracy and good application prospect.
Drawings
Fig. 1 is a schematic flow chart of a deep learning substance decomposition method based on a physical model according to an embodiment of the application.
FIG. 2 is a flow chart illustrating a method for calibrating physical parameters of a CT detector according to an embodiment of the present application.
Fig. 3 is a flowchart illustrating a training method of a pre-trained neural network according to an embodiment of the present application.
FIG. 4 is a schematic diagram of an initial energy spectrum according to an embodiment of the present application.
FIG. 5 is a schematic diagram showing the attenuation spectrum after the X-ray and the substance act in an embodiment of the present application.
FIG. 6 is a schematic diagram of the energy spectrum after the interaction of the X-ray with the detector in an embodiment of the present application.
FIG. 7 is a schematic diagram of the entire probe response process in an embodiment of the present application.
FIG. 8 is a schematic diagram of a detector signal over a period of time in an embodiment of the present application.
Fig. 9 is a schematic diagram of an ASIC internal structure according to an embodiment of the present application.
FIG. 10 is a graph showing response spectra in an embodiment of the present application.
FIG. 11 is a diagram showing the initial energy spectrum to response energy spectrum variation in an embodiment of the present application.
Fig. 12 is a schematic diagram of a first neural network according to an embodiment of the disclosure.
Fig. 13 is a schematic diagram showing a fitting result of the first neural network in an embodiment of the present application.
Fig. 14 is a schematic diagram of a second neural network according to an embodiment of the present application.
FIG. 15 is a graph showing the fitting results of a second neural network in an embodiment of the present application.
Fig. 16 is a schematic diagram showing the overall structures of the first neural network and the second neural network in an embodiment of the present application.
FIG. 17 is a graph showing the fitting results of the first neural network and the second neural network in an embodiment of the present application.
FIG. 18 is a schematic diagram showing the whole process of decomposition of substances in an embodiment of the present application.
Fig. 19 is a schematic structural diagram of an electronic terminal according to an embodiment of the present application.
Fig. 20 is a schematic structural view of a deep learning substance decomposition device based on a physical model according to an embodiment of the present application.
Detailed Description
Other advantages and effects of the present application will become apparent to those skilled in the art from the present disclosure, when the following description of the embodiments is taken in conjunction with the accompanying drawings. The present application may be embodied or carried out in other specific embodiments, and the details of the present application may be modified or changed from various points of view and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments may be combined with each other without conflict.
Furthermore, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context indicates otherwise. It will be further understood that the terms "comprises," "comprising," "includes," and/or "including" specify the presence of stated features, operations, elements, components, items, categories, and/or groups, but do not preclude the presence, presence or addition of one or more other features, operations, elements, components, items, categories, and/or groups. The terms "or" and/or "as used herein are to be construed as inclusive, or meaning any one or any combination. Thus, "A, B or C" or "A, B and/or C" means "any of the following: a, A is as follows; b, a step of preparing a composite material; c, performing operation; a and B; a and C; b and C; A. b and C). An exception to this definition will occur only when a combination of elements, functions or operations are in some way inherently mutually exclusive.
In order to solve the problems in the background art, the invention provides a deep learning substance decomposition method, a device, a terminal and a storage medium based on a physical model, which aim to solve the problems of complex calculation and complex calibration of the existing substance decomposition method. Meanwhile, in order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions in the embodiments of the present invention will be further described in detail by the following examples with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Before explaining the present invention in further detail, terms and terminology involved in the embodiments of the present invention will be explained, and the terms and terminology involved in the embodiments of the present invention are applicable to the following explanation:
<1> computed tomography (Computed Tomography, CT) is an examination of imaging diagnostics. This technique has been referred to as computed axial tomography (Computed Axial Tomography). It uses computer processing combinations of many X-ray measurements made from different angles to generate cross-sectional (tomographic) images of a particular scan area (virtual "slice") of the object so that the user can see the interior of the object without cutting.
<2> photon counting detectors (Photon Counting Detector, PCD) are key to energy spectrum CT imaging technology, which has evolved rapidly over the past decade. The X-ray detection device can detect the attenuation characteristics of X-rays with different energies, breaks the limitation of the traditional X-CT imaging, and is beneficial to qualitative and quantitative analysis of a detection object. The photon counting detector can detect continuous polychromatic X-ray energy spectrums according to different energy intervals respectively, obtain X-ray photons in a specific energy interval through an energy threshold value, and reconstruct CT images containing rich information according to attenuation characteristics of X-rays in different energies.
<3> an artificial Neural Network (english: artificial Neural Network, ANN), abbreviated Neural Network (NN) or Neural-like Network, is a mathematical model or computational model that mimics the structure and function of a biological Neural Network (the central nervous system of an animal, particularly the brain) for estimating or approximating functions in the field of machine learning and cognitive science. Neural networks are calculated from a large number of artificial neuronal junctions. In most cases, the artificial neural network can change the internal structure based on external information, and is an adaptive system, so that the artificial neural network has a learning function in popular sense.
Embodiments of the present invention provide a physical model-based deep learning substance decomposition method, an apparatus of the physical model-based deep learning substance decomposition method, and a storage medium and an electronic terminal storing an executable program for implementing the physical model-based deep learning substance decomposition method. With respect to the implementation of the physical model-based deep learning substance decomposition method, an exemplary implementation scenario of the physical model-based deep learning substance decomposition method will be described in the embodiments of the present invention.
Referring to fig. 1, a flow chart of a method for decomposing a deep learning substance based on a physical model according to an embodiment of the present invention is shown. The deep learning substance decomposition method based on the physical model in the embodiment mainly comprises the following steps:
Step S11: acquiring a response energy spectrum of the CT detector and physical parameters of the CT detector; and the physical parameters of the CT detector are obtained by reducing the error between the experimental response spectrum of the CT detector and the predicted response spectrum of the CT detector through a preset calibration experiment.
Specifically, steps S11 to S12 are used to describe the actual flow of application of the substance decomposing method of the present invention. The response energy spectrum of the CT detector is the final response energy spectrum of the detector chip in actual application and is a known quantity; the physical parameters of the CT detector include, but are not limited to, detector thickness, detection efficiency, detector electric field strength, charge sharing coefficient, chip dead time and chip count threshold, chip threshold drift with count rate, detector energy resolution, which are all listed in this embodiment, and the physical parameters of the CT detector are also known in step S11, and the calibration method is described in steps S21 to S24 below.
Step S12: according to the response energy spectrum of the CT detector and the physical parameters of the CT detector, calculating thickness information of detected substances based on a neural network trained in advance; the pre-trained neural network is used for mapping the functional relation among the physical parameters of the CT detector, the thickness information of the detected substance and the response energy spectrum of the CT detector.
Specifically, the pre-trained neural network in this embodiment can accurately map the functional relationship among the physical parameters of the CT detector, the thickness information of the detected substance, and the response energy spectrum of the CT detector. The construction and training process of the pre-trained neural network is described in steps S31 to S33 below. Therefore, on the premise that the physical parameters of the CT detector and the response spectrum of the CT detector are known, according to the neural network trained in advance, the thickness information of the detected substance can be obtained as another unknown parameter, so that the substance decomposition is completed.
For example, when the neural network fits the functional relationship y=f (x, a, b, c) between the physical parameters of the CT detector, the thickness information of the detected material and the response spectrum of the CT detector, the corresponding x, x is the thickness information of the detected material when the physical parameters a, b, c of the CT detector and the response spectrum y of the CT detector are fixed.
It should be noted that, in the present invention, the application of the neural network is not a traditional input/output method, but uses the characteristic that the neural network can map the nonlinear relationship between any data, so that on the premise of determining the rest of known parameters in the neural network, the value of the rest of unknown parameters can be obtained by back-pushing according to the trained neural network.
As shown in fig. 2, a flow chart of an experiment for calibrating physical parameters of a CT detector in an embodiment of the present application is shown. In some implementations of this embodiment, the physical parameters of the CT detector in step S11 are obtained based on the following calibration experiments:
step S21: acquiring a predicted value of a physical parameter of the CT detector, thickness information of an experimental substance and an experimental response spectrum of the CT detector;
it should be noted that, the initial value of the predicted value of the physical parameter of the CT detector is preferably an empirical value, and the thickness information of the experimental material and the experimental response spectrum of the CT detector are all known parameters in the calibration experiment.
Step S22: inputting the thickness information of the experimental material and the predicted value of the physical parameter of the CT detector into the neural network after the pre-training to obtain a predicted response spectrum of the CT detector;
specifically, the neural network after pre-training can accurately map the functional relation between the physical parameters of the CT detector, the thickness information of the detected substance and the response energy spectrum of the CT detector, so that when the thickness information of the experimental substance and the predicted value of the physical parameters of the CT detector are known, the predicted response energy spectrum of the CT detector output by the neural network can be obtained. The construction and training process of the pre-trained neural network is described in steps S31 to S33 below.
Step S23: calculating an error between a predicted response spectrum of the CT detector and an experimental response spectrum of the CT detector, if the error is greater than a preset error threshold, updating a predicted value of a physical parameter of the CT detector, and repeating the steps until the error is less than the preset error threshold;
specifically, after the predicted response spectrum of the CT detector output by the neural network is obtained, an error between the predicted response spectrum of the CT detector and the experimental response spectrum of the CT detector may be calculated. And if the error between the prediction response spectrum of the CT detector and the experimental response spectrum of the CT detector is larger than the error threshold value, updating the prediction value of the physical parameter of the CT detector through a back propagation algorithm and repeating the steps S21-S23 until the error between the prediction response spectrum of the CT detector and the experimental response spectrum of the CT detector is smaller than the error threshold value.
Step S24: and when the error is smaller than a preset error threshold, making the physical parameter of the CT detector be the predicted value of the physical parameter of the current CT detector.
Specifically, when the error between the predicted response spectrum of the CT detector and the experimental response spectrum of the CT detector is smaller than the error threshold, the predicted value of the physical parameter of the CT detector currently taken reaches the ideal interval, and the predicted value is used as the physical parameter of the CT detector in step S11 for the actual material decomposition process.
FIG. 3 is a flowchart of a training method for a pre-trained neural network according to one embodiment of the present application. In some implementations of this embodiment, the pre-trained neural network of step S12 is obtained based on the following steps:
step S31: building a physical model, wherein the physical model is used for mapping a functional relation among physical parameters of the CT detector, thickness information of detected substances and response energy spectrum of the CT detector under a fixed initial energy spectrum;
specifically, the physical model can be built by simulation software, and the essence of the physical model is to map the functional relation among the physical parameters of the CT detector, the thickness information of the detected substance and the response energy spectrum of the CT detector under the fixed initial energy spectrum.
In the process from an initial energy spectrum to a response energy spectrum, the X-ray mainly comprises links such as attenuation of the X-ray and a substance, interaction of the X-ray and a detector, charge transmission, chip response and the like through multiple attenuation, and the attenuation principle of the X-ray and a corresponding physical model design method in each link are described as follows:
1) Attenuation model of X-ray and substance
After the X-ray energy spectrum is generated, the X-rays pass through the human body and are attenuated. The X-ray and the substance mainly have the following functions: photoelectric effect (photoelectric effect), compton scattering (Compton scattering), electron pair effect (pair production) and rayleigh scattering (Rayleigh scattering).
Regarding an attenuation model of X-rays and substances, an initial energy spectrum is generated by using Spektr, the energy spectrum of a low-energy part is filtered through low-energy filtering, and the attenuated X-ray energy spectrum is calculated by using substance attenuation coefficients provided by a NISTXCOM database. As shown in fig. 4, shown as an initial energy spectrum; as shown in fig. 5, the attenuation spectrum after the X-rays have acted on the material is shown.
2) Model of interaction of X-rays with detector
After the X-rays attenuate with the material, the X-rays interact with the detector, including photoelectric effects, compton scattering, electron pair effects, and rayleigh scattering.
Regarding the interaction model of X-rays and a detector, a GATE tool is used for simulation, and the actions of the position and energy deposition condition of the X-rays with different energies and the escape, reabsorption, charge sharing and the like of characteristic X-rays after the X-rays are injected into the detector are simulated. As shown in fig. 6, is the energy spectrum after the X-rays interact with the detector.
3) Charge transfer model
When X-rays pass through the substance to the detector, the semiconductor material of the detector converts photons into free electron and free hole pairs. Due to the existence of the strong electric field, electrons and holes can respectively move to the anode and the cathode, and the movement process is the transmission of charges.
The charge may be affected by drift, self-repulsion, thermal diffusion, trapping, etc. during transport.
Throughout the transmission they can be formulated as:
wherein n is f Representing the spatial distribution of free electrons, n t Representing the spatial distribution of the trapped electrons, j n Represents electron flux, D n Represents the electron drift constant, mu n Representing the mobility of the electrons and,
representing the voltage. P is p f Representing the spatial distribution of free holes, p t Representing the spatial distribution of trapped holes, j p Represents hole flux, D p Represents the hole drift constant, μ p Representing hole mobility.
Wherein equations 1, 2, 3 are electron transport equations, and equations 4, 5, 6 are hole transport equations. Equations 1 and 4 represent the change over time of the spatial distribution of free electrons and free holes, with the time rate of change of the spatial distribution of free electrons being equal to the divergence of the electron flux minus the capture coefficient of the electrons. Equation 3 and equation 6 show the spatial distribution of the trapped electrons over time, and the trapping rate of electrons increases with time. Equation 2 and equation 5 represent electron flux and hole flux. The electron flux consists of two parts, namely the thermal drift and diffusion of electrons under an electric field.
Regarding the charge transfer model, since the calculation of formulas 1-6 is time consuming, the present invention approximates with an empirical formula to calculate the collected charge (collecting charge) q at the anode side ie As an example.
Wherein Q is pair Is the electron-hole pair charge quantity sigma generated in the whole detector response process t Is the electron cloud size, σ, when the electron reaches the anode 0 Is the initial charge cloud size, D n Is the electron drift coefficient.
Electrons to chargeThe cloud approximation is seen as a two-dimensional gaussian distribution whose standard deviation varies over time. Equation 8 shows the distribution of electron cloud, equation 9 shows the variation of standard deviation with time, equation 10 shows the amount of charge, equation 11 shows the initial standard deviation, equation 12 shows the approximate time for the charge to reach the anode, equation 7 shows the integration of space, and the collected charge q of the anode can be obtained ie 。
4) Signal generation model
During the movement of charge to both ends of the electrode, a signal is generated. Based on the Shockley-Ramo theory, a weighted potential (weighting potential) is calculated, and the weighted potential is combined with the charge to obtain a response signal of the detector.
In the Shockley-Ramon principle, a signal generated by an electron cloud is detected by an electrode of a detector, and the generation of the signal can be divided into two parts. Taking electrons as an example, the charge is divided into two types, the first is charge q, which moves directly to the anode c The second type is an induced charge q generated during electron transport i ：
q t ＝q c +q i (equation 13)
q i ＝∫∫∫ V qψd 3 r (equation 15)
Equation 13 represents a total charge of q i And q c And (3) summing. Equation 14 represents the charge moving to the anode and equation 15 represents the induced charge during transport.
Where ψ represents the weighting potential (weighting potential). Equation 17 shows that at a particular anode, its weighted potential is 1 and the rest are 0. Equation 16 shows that the second derivative of the weighted potential is 0.
Because the above formula is too complex, the invention adopts the following empirical formula:
based on the above empirical formulas 18 to 23, the calculation amount can be greatly reduced and the calculation result error is small. Based on the combination of the Shockley-Ramo theorem, the weight potential psi (x, y, z) of any position of the detector can be calculated.
The charge generated by each of the anode and cathode can be calculated by combining the Shockley-Ramo theorem and the charge transport model described in 3).
Induced charge Q generated by electrons at time t e The method comprises the following steps:
Q e ＝q ie ψ (x (t), y (t), z (t)) (equation 24)
Since the size of the detector pixel is much larger than the size of one electron, the formula (formula 24) can be simplified as:
Q e ＝q ie ψ(0，0，z(t)) (equation 25)
Induced charge Q generated by holes at time t h The method comprises the following steps:
Q h ＝q ih ψ (x, y, z (t)) (equation 26)
Since the size of the holes is much larger than the size of the electrons, a similar approximation cannot be made here. However, for ease of calculation, x, y may be considered as approximations.
By combining equations (equation 25) and (equation 26), the total induced charge Q of the anode can be obtained s ：
Q s (t)＝Q e (t)-Q h (t)＝q ie ψ(0，0，z e (t))-q ih ψ(x，y，z h (t)) (equation 27)
z e (t)＝z 0 -E f μ e t (formula 28)
z h (t)＝z 0 +E f μ h t (formula 29)
Wherein z is e (t) and z h (t) represents the positions of the free electrons and the free charges in the z direction, mu e Represents the mobility of electrons, mu h Representing the mobility of holes, E f Representing a strong electric field.
After the induction charge is obtained, the induction charge is subjected to time derivation, and the induction current I can be obtained i (t)：
And the induced current is the signal obtained by the detector.
Regarding the signal generation model, the present invention performs the following simulation steps: the weighted potential for each position of the detector is calculated in advance according to the Shockley-Ramo theorem. The respective positions of the electrons and holes, and the induced charges generated at this position, are then calculated separately. Multiplying each of them by a weighted potential yields the total detector induced charge. The induced current can be obtained by deriving the time of the obtained induced electric charge of the detector, which is the final response signal of the detector. The entire probe response process is shown in fig. 7. As shown in fig. 8, a detector signal is generated over a period of time.
5) Chip response model
After the probe generates a response signal, the signal enters the chip module. The chip analyzes and counts the signals to finally obtain the final count of photon counting CT. The chip response step comprises current integration, high-pass filtering, time sequence judgment, difference calculation and comparison counting.
As shown in fig. 9, a schematic diagram of the internal structure of an ASIC is shown. The chip (ASIC) is composed of several modules including current integration, high-pass filtering, time sequence judgment, difference calculation and comparison counting. After the detector generates the signal, the ASIC passes the signal through a pre-amplifier (pre-amplifier) to obtain a voltage signal. This voltage is then high pass filtered and only signals above a certain threshold can pass, thus removing noise. If the ASIC is in an active state, the ASIC counts during the energy interval in which the signal is located, thereby completing one cycle of the ASIC. The chip then has a paralyzed state (paralyzed) lasting about 16ns, which is done in order to prevent saturation of the chip with respect to the signal. During paralysis, all signals arriving at the ASIC are not counted, which can lead to distortion of the spectrum. In addition, when two beams arrive at the ASIC at the same time, it is difficult for the ASIC to distinguish them, which can be seen as one beam, which can lead to a stacking effect of the spectra. In addition, the determination of the threshold value may also have some error, which may also lead to errors in the final count of the energy spectrum.
When the induced current of the detector passes through the current integration module, the current is integrated, and a voltage signal is obtained through output:
wherein I (t) is the induced current of the detector, C f Is the feedback capacitance value.
And performing high-pass filtering on the obtained voltage signal.
Let it be at T 0 At the moment, the high-pass filtered signal exceeds the set threshold voltage, and the effective signal of the voltage signal at this moment is V 0 . T is defined herein 1 ，T 2 ，V 2 ：
T 1 ＝T 0 +ΔT deadtime (formula 32)
T 2 ＝T 1 +ΔT reset ＝T 0 +ΔT deadtime +ΔT reset (equation 33)
V 1 ＝V(T 1 ) (equation 34)
ΔV＝V 1 -V 0 (equation 35)
Where Δv will be used for photon energy comparison.
Before the ASIC actually starts to operate, calibration operations are performed on the ASIC. Photons of different energy levels and threshold voltage V are calculated threshold Relationship between them.
After the energy calibration is completed, the theoretical voltage of the photons at each energy can be obtained. Based on this theoretical voltage, the ASIC may classify the resulting Δv, e.g., Δv is in the voltage energy interval corresponding to 30keV photons, where the count is incremented by one.
Thus, the signal accumulation model is built based on the active reset circuit model, and the photon counting is carried out on the X-rays in different energy intervals.
The chip response process is simulated to obtain a chip response spectrum as shown in fig. 10.
As shown in fig. 11, a schematic diagram of the initial spectrum to response spectrum variation is shown. It can be seen that the entire spectrum has changed significantly from the initial spectrum to the final chip response spectrum. There is a need for a method for mapping the relationship between different material thicknesses, different charge sharing coefficients, different chip dead times, different chip count thresholds, initial spectra and response spectra.
For the whole physical model from the initial energy spectrum to the final response spectrum of the detector chip, the following steps can be taken:
S 3 (E)＝DS 2 (E) (equation 38)
S 4 (E)＝f 1 (S 3 (E) Sigma) (equation 39)
S 5 (E)＝f 2 (S 4 (E)，Nτ，Q trigger ) (equation 40)
Wherein S is 0 For initial energy spectrum, S 1 S is the energy spectrum after low energy filtering 2 S is the energy spectrum after the material is attenuated 3 S is the energy spectrum of the ideal response of the detector 4 For the energy spectrum of the real response of the detector, S 5 For ASIC response spectrum, N represents the total count, μ a 、μ b Represents the attenuation coefficient of low-energy filtering of X-rays, l a 、l b Represents the thickness of the filter material, mu 1 、μ 2 Represents the attenuation coefficient of two substances, l 1 、l 2 Representing the thickness of two substances, D represents the theoretical response matrix of the detector, sigma represents the charge sharing coefficient, f i (:) represents the neural network, ΔE represents the shift in the energy spectrum, τ represents the detector dead time (dead time), Q trigger Representing the chip count threshold, N i ' means the detector count at the ith energy interval, E i Indicating the start energy of the i-th energy interval.
Wherein, formula 36 represents the initial energy spectrum filtered by low energy, formula 37 represents the energy spectrum attenuated by the human body, formula 38 represents the ideal response of the detector, formula 39 represents the fit of the neural network to the non-ideal case, formula 40 represents the neural network analog chip response, and formula 41 represents the actual count of the detector. From equations 36 through 41, a deep learning based model can be built to fit the detector response.
In some implementations of this embodiment, the physical model includes a first physical model and a second physical model; the first physical model is used for mapping a functional relation among an initial energy spectrum, detector thickness, detection efficiency, electric field intensity, charge sharing coefficient, thickness information of detected substances and an intermediate energy spectrum of the CT detector, and the second physical model is used for mapping a functional relation among the initial energy spectrum, chip dead time, chip counting threshold, chip threshold drift coefficient with counting rate, energy resolution, thickness information of detected substances and a final response energy spectrum of the CT detector.
Specifically, the physical model may be composed of a first physical model and a second physical model, where the first physical model is a detector response model, and is composed of the foregoing 1) an attenuation model of X-rays and substances, 2) an interaction model of X-rays and detectors, 3) a charge transmission model, and 4) a signal generation model, and is used for mapping a functional relationship between an initial energy spectrum, a thickness of a detector, a detection efficiency, an electric field strength, a charge sharing coefficient, thickness information of detected substances, and an intermediate energy spectrum of the CT detector; the second physical model, namely a chip response model, corresponds to the aforementioned 5) chip response model, and is used for mapping the functional relationship among the initial energy spectrum, the chip dead time, the chip count threshold, the chip threshold drift coefficient with the count rate, the energy resolution, the thickness information of the detected substance and the final response spectrum of the CT detector.
Step S32: generating a data set according to the physical model, wherein the data set comprises physical parameters of different CT detectors under a fixed initial energy spectrum, thickness information of different detected substances and response energy spectrums of the CT detectors corresponding to the physical parameters and the thickness information;
based on the physical model, a large number of data sets can be generated, wherein the data sets comprise physical parameters of different CT detectors under the fixed initial energy spectrum, thickness information of different detected substances and response energy spectrums of the CT detectors corresponding to the physical parameters and the thickness information of different detected substances.
In some implementation processes of the embodiment, a first data set and a second data set are generated through a first physical model and a second physical model respectively, wherein the first data set comprises detector response data of a single energy spectrum and detector response data of a full energy spectrum; the second data set includes chip response data for the full energy spectrum. The specific generation process of the data set is as follows in step S33.
Step S33: the data set is input to an untrained neural network for training the untrained neural network to converge.
In some implementations of the present embodiments, the untrained neural network includes an untrained first neural network and an untrained second neural network; when the untrained neural network is trained, the untrained first neural network is input into an ideal energy spectrum, the thickness of the detector, the detection efficiency, the electric field strength and the charge sharing coefficient, and output into a response energy spectrum of the detector under the thickness of the detector, the detection efficiency, the electric field strength and the charge sharing coefficient; the untrained second neural network input is the output of the untrained first neural network, the chip dead time, the chip count threshold, the chip threshold drift coefficient with count rate, the energy resolution, and the output is the actual response spectrum of the chip.
Specifically, the input of the first neural network is an ideal energy spectrum and a charge sharing coefficient sigma, a detector thickness, detection efficiency and detector electric field intensity, and the output is the response of the detector when the charge sharing coefficient sigma, the detector thickness, the detection efficiency and the detector electric field intensity are the same. Briefly, the goal of the first neural network is to map an ideal energy spectrum to a corresponding energy spectrum at a shared charge coefficient given the shared charge coefficient, detector thickness, detection efficiency, and detector electric field strength.
Fig. 12 is a schematic diagram of the first neural network. In the neural network structure, the first neural network adopts three layers of fully-connected neural networks, namely 242×200, 200×200, 200×241, and the activation function uses tanh (x). The loss function uses a mean square error (mean squared error).
In terms of the manufacturing of the first neural network data set, based on the physical model in the step S31, 4000 groups of detector response data of the single energy spectrum and 4720 groups of detector response data of the full energy spectrum are generated; each set of data simulates the interaction of a particle with a detector (over 100 tens of thousands). There are 6104 sets of data as training data, 1308 sets of data as verification data, 1308 sets of data as test data.
In terms of model performance of the first neural network, the results of the fitting of the first neural network to the probe response are shown in the following table.
List one
As shown in fig. 13, a schematic diagram of the fitting result of the first neural network to the probe response is shown. Wherein (a) and (b) are fitting results of the neural network to the full energy spectrum, and (c) is fitting results of the neural network to the single energy spectrum. The uppermost energy spectrum of (a), (b) and (c) is the input energy spectrum, the middle is the neural network output energy spectrum, and the lowermost is the ideal energy spectrum. As can be seen from the middle of the table I and the figure 13, the neural network model can be used for well fitting the detector response physical model
The input of the second neural network is the output energy spectrum of the first neural network detector response model, the detector dead time tau and the chip counting threshold Q trigger The chip threshold value is output as the response of the chip at the moment according to the counting rate drift coefficient and the energy resolution of the detector. Briefly, the goal of the second neural network is to give τ and Q trigger Under the conditions of the chip threshold value drift coefficient along with the counting rate and the energy resolution of the detector, the ideal energy spectrum is mapped to the corresponding energy spectrum at the moment.
Fig. 14 is a schematic diagram showing the structure of the second neural network. In the neural network structure, the second neural network adopts four layers of fully connected neural networks, namely 243×300, 300×300 and 300×241, and the activation function uses tanh (x). The loss function uses a mean square error (mean squared error).
In terms of the production of the second neural network dataset, based on the physical model in step S31, chip response data of the 1245 set of full energy spectrum is generated. The dataset simulates different X-ray operating current magnitudes (10 mA,50mA, and 200 mA), different material thicknesses, different die dead times (13 ns to 21 ns), different die count thresholds (16 keV to 24 keV). There are 873 sets of data as training data, 186 sets of data as verification data, and 186 sets of data as test data.
In terms of model performance of the second neural network, the results of the fitting of the second neural network to the probe response are shown in the following table.
Watch II
As shown in fig. 15, a schematic diagram of the fitting result of the second neural network to the chip response is shown. And (b) is a fitting result of the neural network to the full energy spectrum. (a) The neutral network output spectrum is used as the input spectrum, and the ideal spectrum is used as the (c). From table two and fig. 15, it can be found that the second neural network can make a good fit to the chip response physical model.
It should be noted that, in this embodiment, a plurality of physical models and a plurality of neural networks are used to map nonlinear relationships among different physical parameters, which is because each physical parameter of the CT detector is set up to a neural network, so that the neural network can learn the characteristics and the relations among the physical parameters more accurately and efficiently, and because the principle that the energy spectrum changes due to each physical parameter has differences, the nonlinear relationships among the thickness information of the substance, each physical parameter of the CT detector and the response energy spectrum are mapped by training and integrating the plurality of neural networks, so that the generalization capability of the neural network can be enhanced, and the internal relations among the parameters can be learned more accurately.
Fig. 16 is a schematic diagram showing the overall structure of the first neural network and the second neural network. It should be noted that, in the figure, the super parameter set 1 is the thickness of the detector, the detection efficiency, the electric field intensity of the detector, and the charge sharing coefficient; super parameter group 2, namely chip dead time, chip count threshold, chip threshold drift with count rate, detector energy resolution; the input energy spectrum is the initial energy spectrum of the CT detector; the output energy spectrum is the final response spectrum of the CT detector, and the concepts have one-to-one correspondence and are not ambiguous.
The loss function expression of the first neural network is as follows:
the loss function expression of the second neural network is:
as shown in fig. 17, the fitting effect of the neural network as a whole to the physical model is shown. The fitting results are shown in fig. 18, (a) is the predicted result of the neural network, and (b) is the system true value. As shown in the figure, the overall fitting result of the neural network basically accords with the expectation, and useful information can be well grasped.
In summary, the complete flow of the present application may be as shown in fig. 18.
Step S181: and constructing a physical model for mapping the functional relation among the initial energy spectrum, the physical parameters, the thickness information of the detected substance and the response energy spectrum of the CT detector.
Specifically, a full-flow simulation is performed on the photon counting detector, and the simulation can simulate phenomena such as charge sharing, signal stacking and the like. After the simulation program is completed, a large amount of simulation is performed to obtain different charge sharing coefficients, different detector dead times and different response results of the system when different chips start to be powered.
Step S182: a dataset is generated based on the physical model, and a neural network is constructed and trained to fit the physical model.
The neural network introduces multiple physical parameters simultaneously: the detector thickness, detection efficiency, electric field strength, charge sharing coefficient, chip dead time, chip count threshold, chip threshold drift coefficient with count rate, energy resolution, and deep learning is used to fit the response of the whole system.
Step S183: and setting a plurality of calibration experiments, and determining physical parameters of the CT detector by reducing errors of the predicted value and the experimental response value of the neural network.
After the neural network model can obtain a relatively accurate fitting result, the physical parameters of the CT detector can be reversely deduced by fixing the neural network parameters and the thickness information of the detected substances through a plurality of groups of calibration experiments.
Step S184: and in the imaging process, calculating thickness information of the detected substance based on the neural network according to physical parameters of the CT detector and actual response of the CT detector.
After the physical parameters of the CT detector are determined and the neural network is trained, the most conforming material thickness information can be obtained based on the physical parameters of the CT detector, the neural network and the actual response energy spectrum, and material decomposition is completed.
The method has the advantages compared with the traditional substance decomposition method that: a large number of free variables need to be fitted in the traditional method such as a calibration method, and only a small number of physical parameters need to be fitted after the neural network parameters are determined in the method; the traditional method has extremely complex calibration operation, and the method can determine physical parameters only by a plurality of times of calibration operation; the traditional method such as the model method has complex model calculation, and the method can obtain the thickness information of the substance without complex operation.
Although the steps are described in the above-described sequential order in the above-described embodiments, it will be appreciated by those skilled in the art that in order to achieve the effects of the present embodiments, the steps need not be performed in such order, and may be performed simultaneously (in parallel) or in reverse order, and such simple variations are within the scope of the present invention.
Referring to fig. 19, for a hardware structure of a physical model-based deep learning substance decomposition terminal, an optional hardware structure diagram of a physical model-based deep learning substance decomposition terminal 1900 according to an embodiment of the present invention may be shown, where the terminal 1900 may be a mobile phone, a computer device, a tablet device, a personal digital processing device, a factory background processing device, or the like. The deep learning substance decomposition terminal 1900 based on the physical model includes: at least one processor 1901, memory 1902, at least one network interface 1904, and a user interface 1906. The various components in the device are coupled together by a bus system 1905. It is to be appreciated that the bus system 1905 is employed to facilitate connection communications between these components. The bus system 1905 includes a power bus, a control bus, and a status signal bus in addition to the data bus. But for clarity of illustration the various buses are labeled as bus systems in fig. 19.
The user interface 1906 may include, among other things, a display, keyboard, mouse, trackball, click gun, keys, buttons, touch pad, or touch screen, etc.
It is to be appreciated that memory 1902 can be either volatile memory or nonvolatile memory, and can include both volatile and nonvolatile memory. The nonvolatile Memory may be a Read Only Memory (ROM), a programmable Read Only Memory (PROM, programmable Read-Only Memory), which serves as an external cache, among others. By way of example and not limitation, many forms of RAM are available, 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 by embodiments of the present invention is intended to comprise, without being limited to, these and any other suitable types of memory.
The memory 1902 in an embodiment of the present invention is configured to store various categories of data to support the operation of the deep learning substance decomposition terminal 1900 based on a physical model. Examples of such data include: any executable programs for operating on the physical model-based deep learning substance decomposition terminal 1900, such as an operating system 19021 and application programs 19022; the operating system 19021 contains various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and handling hardware-based tasks. The application program 19022 may include various application programs such as a Media Player (Media Player), a Browser (Browser), etc. for implementing various application services. The deep learning substance decomposition method based on the physical model provided by the embodiment of the invention can be contained in the application program 19022.
The method disclosed in the above embodiment of the present invention may be applied to the processor 1901 or implemented by the processor 1901. The processor 1901 may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the methods described above may be performed by integrated logic circuitry in hardware or instructions in software in the processor 1901. The processor 1901 may be a general purpose processor, a digital signal processor (DSP, digital Signal Processor), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like. The processor 1901 may implement or perform the methods, steps, and logic blocks disclosed in embodiments of the present invention. The general purpose processor 1901 may be a microprocessor or any conventional processor or the like. The steps of the accessory optimization method provided by the embodiment of the invention can be directly embodied as the execution completion of the hardware decoding processor or the execution completion of the hardware and software module combination execution in the decoding processor. The software modules may be located in a storage medium having memory and a processor reading information from the memory and performing the steps of the method in combination with hardware.
In an exemplary embodiment, the physical model-based deep learning substance decomposition terminal 1900 may be implemented by one or more application specific integrated circuits (ASICs, application Specific Integrated Circuit), DSPs, programmable logic devices (PLDs, programmable Logic Device), complex programmable logic devices (CPLDs, complex Programmable Logic Device) for performing the aforementioned methods.
As shown in fig. 20, a schematic structural diagram of a deep learning substance decomposing device based on a physical model in an embodiment of the present invention is shown. In the present embodiment, the deep learning substance decomposition device 2000 based on the physical model includes: parameter acquisition module 2001: the method comprises the steps of acquiring a response energy spectrum of a CT detector and physical parameters of the CT detector; the physical parameters of the CT detector are obtained by reducing the error between an experimental response spectrum of the CT detector and a predicted response spectrum of the CT detector through a preset calibration experiment; thickness calculation module 2002: according to the response energy spectrum of the CT detector and the physical parameters of the CT detector, calculating thickness information of detected substances based on a neural network trained in advance; the pre-trained neural network is used for mapping the functional relation among the physical parameters of the CT detector, the thickness information of the detected substance and the response energy spectrum of the CT detector.
It should be noted that: the physical model-based deep learning substance decomposition device provided in the above embodiment is only exemplified by the division of the above program modules when performing the physical model-based deep learning substance decomposition, and in practical application, the above processing allocation may be performed by different program modules according to needs, that is, the internal structure of the device is divided into different program modules, so as to complete all or part of the above processing. In addition, the deep learning substance decomposition device based on the physical model provided in the above embodiment belongs to the same concept as the deep learning substance decomposition method based on the physical model, and the specific implementation process thereof is detailed in the method embodiment, which is not described herein again.
Those of ordinary skill in the art will appreciate that: all or part of the steps for implementing the method embodiments described above may be performed by computer program related hardware. The aforementioned computer program may be stored in a computer readable storage medium. The program, when executed, performs steps including the method embodiments described above; and the aforementioned storage medium includes: various media that can store program code, such as ROM, RAM, magnetic or optical disks.
In the embodiments provided herein, the computer-readable storage medium may include read-only memory, random-access memory, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, flash memory, U-disk, removable hard disk, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer. In addition, any connection is properly termed a computer-readable medium. For example, if the instructions are transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital Subscriber Line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. It should be understood, however, that computer-readable and data storage media do not include connections, carrier waves, signals, or other transitory media, but are intended to be directed to non-transitory, tangible storage media. Disk and disc, as used herein, includes Compact Disc (CD), laser disc, optical disc, digital Versatile Disc (DVD), floppy disk and blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers.
In summary, the present application provides a method, an apparatus, a terminal, and a medium for decomposing a deep learning substance based on a physical model, and the present invention provides a method for improving the decomposition efficiency of a deep learning substance based on a physical model, which is used for solving the problems of complex calculation and complex calibration of the existing substance decomposition method. Therefore, the method effectively overcomes various defects in the prior art and has high industrial utilization value.
The foregoing embodiments are merely illustrative of the principles of the present application and their effectiveness, and are not intended to limit the application. Modifications and variations may be made to the above-described embodiments by those of ordinary skill in the art without departing from the spirit and scope of the present application. Accordingly, it is intended that all equivalent modifications and variations which may be accomplished by persons skilled in the art without departing from the spirit and technical spirit of the disclosure be covered by the claims of this application.
Claims (10)
1. A physical model-based deep learning substance decomposition method, characterized by comprising:
acquiring a response energy spectrum of the CT detector and physical parameters of the CT detector; the physical parameters of the CT detector are obtained by reducing the error between an experimental response spectrum of the CT detector and a predicted response spectrum of the CT detector through a preset calibration experiment;
According to the response energy spectrum of the CT detector and the physical parameters of the CT detector, calculating thickness information of detected substances based on a neural network trained in advance; the pre-trained neural network is used for fitting a physical model, and the physical model is used for mapping a functional relation among an initial energy spectrum, physical parameters of a CT detector, thickness information of detected substances and a response energy spectrum of the CT detector.
2. The physical model-based deep learning substance decomposition method of claim 1, wherein the physical parameters of the CT detector are derived based on the following calibration experiments:
acquiring a predicted value of a physical parameter of the CT detector, thickness information of an experimental substance and an experimental response spectrum of the CT detector;
inputting the thickness information of the experimental material and the predicted value of the physical parameter of the CT detector into the neural network after the pre-training to obtain a predicted response spectrum of the CT detector;
calculating an error between a predicted response spectrum of the CT detector and an experimental response spectrum of the CT detector, if the error is greater than a preset error threshold, updating a predicted value of a physical parameter of the CT detector, and repeating the steps until the error is less than the preset error threshold;
And when the error is smaller than a preset error threshold, enabling the physical parameter of the CT detector to be equal to the predicted value of the physical parameter of the current CT detector.
3. The physical model-based deep learning substance decomposition method of claim 1, wherein the pre-trained neural network is obtained based on the steps of:
building a physical model, wherein the physical model is used for mapping a functional relation among an initial energy spectrum, physical parameters of a CT detector, thickness information of detected substances and a response energy spectrum of the CT detector;
generating a data set according to the physical model, wherein the data set comprises physical parameters of different CT detectors under a fixed initial energy spectrum, thickness information of different detected substances and response energy spectrums of the CT detectors corresponding to the physical parameters and the thickness information;
the data set is input to an untrained neural network for training the untrained neural network to converge.
4. A method of deep learning material decomposition based on physical model according to any of claims 1-3, wherein the physical parameters of the CT detector include detector thickness, detection efficiency, electric field strength, charge sharing coefficient, chip dead time, chip count threshold, chip threshold drift with count rate coefficient, energy resolution.
5. The physical model-based deep learning substance decomposition method of claim 4, wherein said pre-trained neural network comprises a pre-trained first neural network and a pre-trained second neural network; the pre-trained first neural network is used for mapping a functional relation among detector thickness, detection efficiency, electric field intensity, charge sharing coefficient, thickness information of detected substances and an intermediate energy spectrum of the CT detector, and the pre-trained second neural network is used for mapping a functional relation among chip dead time, chip counting threshold, chip threshold drift coefficient with counting rate, energy resolution, thickness information of detected substances and a final response energy spectrum of the CT detector.
6. The physical model-based deep learning substance decomposition method of claim 4, wherein said physical model comprises a first physical model and a second physical model; the first physical model is used for mapping a functional relation among an initial energy spectrum, detector thickness, detection efficiency, electric field intensity, charge sharing coefficient, thickness information of detected substances and an intermediate energy spectrum of the CT detector, and the second physical model is used for mapping a functional relation among the initial energy spectrum, chip dead time, chip counting threshold, chip threshold drift coefficient with counting rate, energy resolution, thickness information of detected substances and a final response energy spectrum of the CT detector.
7. The physical model-based deep learning substance decomposition method of claim 4, wherein said untrained neural network comprises an untrained first neural network and an untrained second neural network; when the untrained neural network is trained, the untrained first neural network is input into an ideal energy spectrum, the thickness of the detector, the detection efficiency, the electric field strength and the charge sharing coefficient, and output into a response energy spectrum of the detector under the thickness of the detector, the detection efficiency, the electric field strength and the charge sharing coefficient; the untrained second neural network input is the output of the untrained first neural network, the chip dead time, the chip count threshold, the chip threshold drift coefficient with count rate, the energy resolution, and the output is the actual response spectrum of the chip.
8. A deep learning substance decomposition device based on a physical model, characterized by comprising:
parameter acquisition module: the method comprises the steps of acquiring a response energy spectrum of a CT detector and physical parameters of the CT detector; the physical parameters of the CT detector are obtained by reducing the error between an experimental response spectrum of the CT detector and a predicted response spectrum of the CT detector through a preset calibration experiment;
And a thickness calculation module: the thickness information of the detected substance is calculated based on the neural network trained in advance according to the response energy spectrum of the CT detector and the physical parameters of the CT detector; the pre-trained neural network is used for fitting a physical model, and the physical model is used for mapping a functional relation among an initial energy spectrum, physical parameters of a CT detector, thickness information of detected substances and a response energy spectrum of the CT detector.
9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the method of any of claims 1 to 7.
10. An electronic terminal, comprising: a processor and a memory;
the memory is used for storing a computer program;
the processor is configured to execute the computer program stored in the memory, to cause the terminal to perform the method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211711964.8A CN116011327A (en) | 2022-12-29 | 2022-12-29 | Deep learning substance decomposition method, device, terminal and storage medium based on physical model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211711964.8A CN116011327A (en) | 2022-12-29 | 2022-12-29 | Deep learning substance decomposition method, device, terminal and storage medium based on physical model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116011327A true CN116011327A (en) | 2023-04-25 |
Family
ID=86026386
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211711964.8A Pending CN116011327A (en) | 2022-12-29 | 2022-12-29 | Deep learning substance decomposition method, device, terminal and storage medium based on physical model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116011327A (en) |
-
2022
- 2022-12-29 CN CN202211711964.8A patent/CN116011327A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110097611B (en) | Image reconstruction method, device, equipment and storage medium | |
| CN107229787B (en) | Gamma energy spectrum analysis method based on approximation coefficient and deep learning | |
| US11686863B2 (en) | Neural network-based corrector for photon counting detectors | |
| CN110731773B (en) | Abnormal electrocardiogram screening method based on fusion of global and local depth features of electrocardiogram | |
| CN109242920B (en) | Method, device and system for decomposing substance | |
| Saadi et al. | ABC optimized neural network model for image deblurring with its FPGA implementation | |
| JP7143330B2 (en) | Method and apparatus for image reconstruction | |
| CN110415307A (en) | A kind of multipotency CT imaging method based on tensor completion, device and its storage equipment | |
| Gagne et al. | Signal detectability in digital radiography: spatial domain figures of merit | |
| US20220252744A1 (en) | Method and device for identifying atomic species emitting x- or gamma radiation | |
| Feng et al. | Neural-networks-based photon-counting data correction: pulse pileup effect | |
| Liu et al. | Recovery of saturated signal waveform acquired from high-energy particles with artificial neural networks | |
| CN116432703B (en) | Pulse height estimation method, system and terminal based on composite neural network model | |
| Aliaga et al. | Corrected position estimation in PET detector modules with multi-anode PMTs using neural networks | |
| CN116011327A (en) | Deep learning substance decomposition method, device, terminal and storage medium based on physical model | |
| Ma et al. | X-ray spectra correction based on deep learning CNN-LSTM model | |
| CN110675467B (en) | Material decomposition method and device for energy spectrum CT, storage medium and electronic equipment | |
| Sharma et al. | Prediction of seismicity cycles in the Himalayas using artificial neural network | |
| CN116152373A (en) | Low-dose CT image reconstruction method combining neural network and convolutional dictionary learning | |
| Rezaei et al. | Numerical evaluation of gamma radiation monitoring | |
| Cennini et al. | A neural network approach for the TPC signal processing | |
| CN113807421B (en) | Feature map processing method of attention module based on pulse sending cortex model | |
| CN110687584A (en) | Fast nuclide identification method based on LSTM | |
| CN114916950B (en) | High-spatial-resolution energy spectrum CT image reconstruction method based on multilayer flat panel detector | |
| CN110221338B (en) | Method for reconstructing radiation field in peripheral area of shield |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |