阅读说明：本技术 一种复合能谱ct成像方法 (Composite energy spectrum CT imaging method ) 是由 曾凯 徐丹 于 2019-10-08 设计创作，主要内容包括：本发明提出一种复合能谱CT成像方法,该方法基于传统CT系统,扫描过程中通过高压发生器按照指定频率切换低电压和高电压,获得复合扫描数据；对低电压扫描数据和高电压扫描数据分别进行图像重建,得到低电压图像和高电压图像；对重建后的低电压图像和高电压图像进行图像域的物质分解得到基物质图像。本发明基于传统CT系统实现,电压切换的速度可以根据现有硬件的条件来调整,不需要对硬件进行任何改动,因此实现成本极低。由于切换速度的降低,电压能保持和单能扫描一样的稳定性,因此不需要进行特殊的校准。相对于GE的高速切换方式,本发明主要有成本低,易实现,能量分离大,剂量低的优点。(The invention provides a composite energy spectrum CT imaging method, which is based on a traditional CT system, wherein in the scanning process, a high-voltage generator switches low voltage and high voltage according to specified frequency to obtain composite scanning data; respectively carrying out image reconstruction on the low-voltage scanning data and the high-voltage scanning data to obtain a low-voltage image and a high-voltage image; and performing image domain material decomposition on the reconstructed low-voltage image and the reconstructed high-voltage image to obtain a base material image. The invention is realized based on the traditional CT system, the voltage switching speed can be adjusted according to the conditions of the existing hardware, and the hardware does not need to be changed, so the realization cost is extremely low. Due to the reduced switching speed, the voltage can maintain the same stability as a single energy scan, and therefore no special calibration is required. Compared with a GE high-speed switching mode, the method mainly has the advantages of low cost, easiness in implementation, large energy separation and low dosage.)

1. A method of composite spectral CT imaging comprising the steps of:

(1) in the scanning process, the low voltage and the high voltage are switched by the high voltage generator according to the designated frequency to obtain composite scanning data, and the method specifically comprises the following steps of performing any one of the switching modes of a, b and c in one scanning period:

a. the high voltage and the low voltage are switched once every time the ray source rotates one circle;

b. the high-low voltage is switched once when the ray source rotates for N circles, N is more than or equal to 0.5 and less than 1, and at least half circle of high-voltage scanning data and half circle of low-voltage scanning data are obtained in one scanning period respectively, and the ray is closed or not closed between the previous high-voltage/low-voltage scanning and the next low-voltage/high-voltage scanning;

c. switching is carried out for multiple times in each circle of scanning process, voltage switching modes in two adjacent circles of scanning processes are complementary, high-voltage scanning data areas and low-voltage scanning data areas in each circle of scanning data are alternately distributed, and in two adjacent circles of scanning data, the high-voltage scanning data areas in the previous circle correspond to the low-voltage scanning data areas in the next circle in position one by one;

(2) respectively carrying out image reconstruction on the low-voltage scanning data and the high-voltage scanning data to obtain a low-voltage image and a high-voltage image;

(3) and performing image domain material decomposition on the reconstructed low-voltage image and the reconstructed high-voltage image to obtain a base material image.

2. The method of claim 1, wherein the scan acquisition scan comprises an axial scan and a helical scan.

3. A method according to claim 2, wherein when using helical scanning, for the switching mode of b, the helical pitch p of the scan satisfies:

4. A method of composite spectral CT imaging according to claim 1, wherein said method of image reconstruction comprises: filtered back projection, iterative reconstruction, and compressive sensing.

5. The composite energy spectrum CT imaging method according to claim 2, wherein when the helical scanning is adopted for scanning, the step (2) adopts an artifact-removed image reconstruction method to respectively perform image reconstruction on the low voltage scanning data and the high voltage scanning data, and the specific steps include:

(51) constructing training data: obtaining a clear artifact-free CT image as a target image; carrying out spiral CT scanning on a phantom corresponding to a target image to obtain spiral scanning conical beam projection data, and carrying out image reconstruction on the spiral scanning conical beam projection data by using the conventional image reconstruction method to obtain an initial image; carrying out a numerical simulation process on the initial image to generate corresponding simulated cone beam projection data; reconstructing the simulated cone beam projection data by adopting the same image reconstruction method as the initial image to obtain a secondary image;

(52) building a convolutional neural network, and adding an input channel to an input layer of the convolutional neural network, namely the built convolutional neural network has two input channels; respectively sending the initial image and the secondary image into two input channels, extracting gray information and structural information of the initial image and the secondary image through a convolutional neural network, and estimating a real image after autonomous learning is carried out according to the extracted gray information and structural information;

(53) constructing a loss function about a target image and an estimated real image, and training the convolutional neural network by using a gradient descent method; the loss function is:

wherein Img denotes an output image of the convolutional neural network, Img_tureRepresenting the target image corresponding to Img, Img_kThe value of the pixel representing the k-th pixel in the image Img, Img_ture,kRepresenting an image Img_tureThe representing pixel value of the kth pixel point;

(54) obtaining an initial image and a secondary image by respectively adopting the method in the step (51) for the low-voltage spiral scanning cone beam projection data and the high-voltage spiral scanning cone beam projection data obtained in the step (1); and sending the obtained initial image and the secondary image into a trained convolutional neural network to obtain a corresponding clear artifact-free high/voltage reconstructed image.

6. The method of claim 5, wherein the convolutional neural network comprises: CNN, ResNet and Unet.

Technical Field

The invention relates to the technical field of energy spectrum imaging, in particular to a composite energy spectrum CT imaging method.

Background

The energy spectrum imaging technology has important significance for medical image diagnosis, and the energy spectrum imaging technology can separate information of different energies of substances, remarkably inhibit ray hardening artifacts and bring more basis for clinical diagnosis. However, to achieve energy spectrum imaging, hardware systems more advanced than the conventional CT system, such as a siemens dual-ray source dual-detector CT system, a general electric high-speed switching CT scanning system, and a philips dual-layer detector CT system, must be adopted. These systems all rely on their own proprietary hardware technology to scan material information for different energies. Therefore, the conventional CT system is difficult to perform spectral imaging in practical clinical applications.

Energy spectrum CT has been widely used in scientific research and clinical practice since Siemens introduced solutions with dual sources and dual detectors in 2006. Conventional CT imaging provides an image of the scanned object's effective absorption coefficient for X-rays at a certain scan kV, depending on the size of the object, the radiation filter used, etc. There are differences in CT values across different CT scanning systems even under the same scanning conditions. Quantitative analysis is therefore relatively difficult on conventional CT. The energy spectrum CT can sense the attenuation condition of the object to X-rays with different energies by scanning the object under different energies, so that the composition of substances in the scanned object can be distinguished, the obtained image is slightly influenced by scanning conditions, and quantitative analysis can be provided more accurately.

Spectral CT achieves material decomposition by measuring the absorption of x-rays of different energies by the object (patient) being scanned. In order to obtain better image quality, generally, this measurement at two or more energies requires good simultaneity, as well as large energy separation. Currently there are several implementations shown in table 1 in the products available on the market: the spectrum fast switching that general electric company adopted, secondly siemens's double-source dual detector, the third is philips's double-deck detector. In addition, photon counting CT is under development, and no product capable of being used clinically exists in the market.

TABLE 1 existing spectral CT technique

The simultaneity and energy separation of photon counting CT is best in the state of the art, but it is also rather expensive to manufacture. Moreover, the stability problem of the photon counting detector under high dose is not solved well, so that the commercialization of the detector is difficult. Although fast spectrum switching has good simultaneity and low cost, its switching speed is also limited. When the switching speed is faster and faster, the difference between the switched energy spectrums is smaller and smaller due to the existence of certain rising and falling time of the bulb voltage, and the effect of reconstructing an image is poor. The cost of dual source dual detectors is also high, so spectral scanning technology has been available only in the top-most CT products of siemens. Although the double-layer detector has good simultaneity, the energy resolving power of the double-layer detector is the worst of all the technologies, and the cost of the detector is high.

Disclosure of Invention

The purpose of the invention is as follows: in order to reduce the hardware requirement of CT imaging on an energy spectrum CT system, the invention provides a composite energy spectrum CT imaging method, which is provided based on the traditional CT system, mainly obtains an energy spectrum image through scanning control, a proper image reconstruction algorithm and an energy spectrum decomposition technology, and does not need to be changed on hardware.

The technical scheme is as follows: in order to achieve the purpose, the technical scheme provided by the invention is as follows:

a method of composite spectral CT imaging comprising the steps of:

(1) in the scanning process, the low voltage and the high voltage are switched by the high voltage generator according to the designated frequency to obtain composite scanning data, and the method specifically comprises the following steps of performing any one of the switching modes of a, b and c in one scanning period:

a. the high voltage and the low voltage are switched once every time the ray source rotates one circle;

b. the high-low voltage is switched once when the ray source rotates for N circles, N is more than or equal to 0.5 and less than 1, and at least half circle of high-voltage scanning data and half circle of low-voltage scanning data are obtained in one scanning period respectively, and the ray is closed or not closed between the previous high-voltage/low-voltage scanning and the next low-voltage/high-voltage scanning;

c. switching is carried out for multiple times in each circle of scanning process, voltage switching modes in two adjacent circles of scanning processes are complementary, high-voltage scanning data areas and low-voltage scanning data areas in each circle of scanning data are alternately distributed, and in two adjacent circles of scanning data, the high-voltage scanning data areas in the previous circle correspond to the low-voltage scanning data areas in the next circle in position one by one;

(2) respectively carrying out image reconstruction on the low-voltage scanning data and the high-voltage scanning data to obtain a low-voltage image and a high-voltage image;

(3) and performing image domain material decomposition on the reconstructed low-voltage image and the reconstructed high-voltage image to obtain a base material image.

Further, the scan acquisition scan includes an axial scan and a helical scan.

Further, when the helical scanning is adopted, for the switching mode described in b, the pitch p of the scanning satisfies:

the ray off time between the previous high/low voltage scan and the next low/high voltage scan is

Further, the image reconstruction method comprises the following steps: filtered back projection, iterative reconstruction, and compressive sensing.

Further, when scanning is performed in a helical scanning manner, image reconstruction is performed on the low-voltage scanning data and the high-voltage scanning data respectively by using an artifact-removed image reconstruction method in the step (2), and the specific steps include:

(51) constructing training data: obtaining a clear artifact-free CT image as a target image; carrying out spiral CT scanning on a phantom corresponding to a target image to obtain spiral scanning conical beam projection data, and carrying out image reconstruction on the spiral scanning conical beam projection data by using the conventional image reconstruction method to obtain an initial image; carrying out a numerical simulation process on the initial image to generate corresponding simulated cone beam projection data; reconstructing the simulated cone beam projection data by adopting the same image reconstruction method as the initial image to obtain a secondary image;

(52) building a convolutional neural network, and adding an input channel to an input layer of the convolutional neural network, namely the built convolutional neural network has two input channels; respectively sending the initial image and the secondary image into two input channels, extracting gray information and structural information of the initial image and the secondary image through a convolutional neural network, and estimating a real image after autonomous learning is carried out according to the extracted gray information and structural information;

(53) constructing a loss function about a target image and an estimated real image, and training the convolutional neural network by using a gradient descent method; the loss function is:

wherein Img denotes an output image of the convolutional neural network, Img_tureRepresenting the target image corresponding to Img, Img_kThe value of the pixel representing the k-th pixel in the image Img, Img_ture,kRepresenting an image Img_tureThe representing pixel value of the kth pixel point;

(54) obtaining an initial image and a secondary image by respectively adopting the method in the step (51) for the low-voltage spiral scanning cone beam projection data and the high-voltage spiral scanning cone beam projection data obtained in the step (1); and sending the obtained initial image and the secondary image into a trained convolutional neural network to obtain a corresponding clear artifact-free high/voltage reconstructed image.

Specifically, the convolutional neural network includes: CNN, ResNet and Unet.

Has the advantages that: compared with the prior art, the invention has the following advantages:

although the dual-energy scanning is realized by high-voltage switching as in the GE, the dual-energy scanning does not need to be switched every sampling as in the GE system. The switching speed can be adjusted according to the conditions of the existing hardware, and the hardware does not need to be changed at all, so the realization cost is extremely low. Due to the reduced switching speed, the voltage can maintain the same stability as a single energy scan, and therefore no special calibration is required. The exposure current is also easily modified and has the ability to achieve automatic current modulation. Generally, compared with the GE high-speed switching mode, the invention mainly has the advantages of low cost, easy realization, large energy separation and low dosage.

Drawings

FIG. 1 is a flow chart of the present invention;

FIG. 2 is a schematic diagram of two complete data loops scanned at low voltage and high voltage respectively when an axial scanning mode is adopted;

fig. 3 is a schematic diagram of scanning data of two adjacent circles obtained by switching high and low voltages for multiple times in the same circle when an axis scanning mode is adopted;

FIG. 4 is a schematic diagram of scanning two half-turn data at high and low voltages, respectively, when an axial scanning mode is employed;

FIG. 5 is a diagram showing scanning data of switching between low voltage and high voltage every other turn in the spiral mode;

FIG. 6 is a diagram showing scan data for switching between low and high voltages every 0.75 turns in the spiral mode;

FIG. 7 is a graph of scan data for a spiral mode with reduced current in a particular direction or with radiation completely turned off;

fig. 8 is a structural diagram of a convolutional neural network.

Detailed Description

The present invention will be further described with reference to the accompanying drawings.

The energy spectrum CT imaging method is provided based on the traditional CT system, the conventional CT system is used for realizing a more flexible voltage switching scanning mode to acquire data, and an energy spectrum image is obtained by utilizing image reconstruction and energy spectrum decomposition technologies. The method does not require any changes in hardware.

For spectral imaging, it is most critical to obtain material information of different energy spectra. The invention provides a method for realizing dual-energy scanning by utilizing the existing CT hardware, switching between low voltage and high voltage on the basis of only changing a scanning control mode to obtain a group of scanning data under composite energy, and then utilizing an advanced reconstruction technology. In one scanning of the CT system, the high-voltage controller is repeatedly switched between low voltage and high voltage at the speed supported by the existing hardware, so that the aim of dual-energy scanning is fulfilled. During the scan, the high voltage generator switches back and forth between a low voltage and a high voltage. The switching frequency can be adjusted to ensure that enough data can be provided to reconstruct images of low voltage and high voltage respectively and independently under each energy.

Scanning typically has an axial scan mode and a helical scan mode.

In the axial scan mode, the patient's bed is stationary and generally at least half a turn of data is required under each voltage in order to have enough data to reconstruct an image at both low and high voltages after exposure. In this mode, various switching modes can be designed.

Fig. 2 to 4 show a schematic view of the scanning data in the 3-axis scanning mode with a scanning period of 2 revolutions. Fig. 2 shows that when the axis scanning mode is adopted, a circle of complete data is scanned at a low voltage and a high voltage respectively. Fig. 3 shows a case where the axis scanning mode is used, and two complete turns of data are scanned, where the switching is performed several times in each turn, but the switching modes of the first turn and the second turn are opposite, that is, the high voltage region in the first turn and the low voltage region in the second turn are overlapped with each other, and the low voltage region in the first turn and the high voltage region in the second turn are overlapped with each other. Fig. 4 is a schematic diagram of scanning two half-turn data at high and low voltages respectively when the axis scanning mode is adopted. In these switching modes, the low voltage and the high voltage respectively have enough data to reconstruct an image.

In the helical scanning mode, the helical pitch and the switching interval can be controlled to ensure that complete images can be respectively reconstructed under low voltage and high voltage. In the helical scan mode, the pitch is p (distance moved/scan width per turn). The period of each scan switch may be defined by 1/p turns, for example, p is 0.5, and each two turns is a scan period. Fig. 5 to 7 show the scanning data in 3 helical scanning modes for each scanning period. FIG. 5 shows the switching of voltage every rotation of the radiation source in the helical mode; FIG. 6 shows the switching of the voltage for each 0.75 rotation of the radiation source in the helical mode; fig. 7 shows the source in helical mode scanning 0.75 turns at high voltage and then switching low voltage for another 0.75 turns, but introducing a time interval (about 0.25 turns) between the previous and the next scan, during which the current is reduced or the radiation is completely turned off.

In the case of FIG. 7, the source is switched between high and low voltages every N revolutions, 0.5 ≦ N<1, only need to guarantee to obtain half a turn of high voltage scanning data and low voltage scanning data respectively at least in a scanning cycle, close or not close the ray between the last high voltage/low voltage scanning and the last low voltage/high voltage scanning. This N can also take other values which satisfy the conditions, for example p is 0.5 and N is 0.75, in which case the switch-off time is the time for switching off the radiation between two successive scans

Here, vof is 0.5 turns; alternatively, p may be 0.75 and N may be 0.75, such that vof is 0.

The axis scanning and spiral dual-energy scanning mode can independently reconstruct images in the whole scanning range from the exposure data with low voltage and high voltage. After obtaining the low voltage and high voltage images, the material decomposition can be performed directly in the image domain.

The technical solution of the present invention is further explained by the following specific implementation, and the flow of the composite energy spectrum CT imaging method proposed by the embodiment is shown in fig. 1, and includes the steps of:

step1 scanning is performed using the existing CT system based on the voltage switching pattern illustrated in fig. 2 to 7. The switching interval of the voltage may be at a speed that can be achieved by the system hardware. For example, the scanning speed is one turn in 0.5 second, and the switching is performed once in 0.75 turn, the low voltage is 80kVp, and the high voltage is 140 kVp.

And Step2, after data are acquired through scanning, respectively correcting and preprocessing the low-voltage data and the high-voltage data, and respectively reconstructing images of the two voltages by utilizing a CT image reconstruction technology.

For axial scan data, the reconstruction algorithm includes but is not limited to: filtering back projection, iterative reconstruction, compressed sensing and the like.

For helical scan data, when reconstructing scan data of low voltage and high voltage respectively, due to the fact that the switching frequency is relatively low and the reconstruction error is proportional to the square of the cone angle (proportional to the row number of the detector), when the row number of the detector is increased to 128 or even 256 rows, a large error is brought, and particularly, a relatively serious artifact exists in a reconstructed image. In order to eliminate the image artifact, there is preferably provided an improved helical reconstruction method for reducing the effect of the cone beam artifact on the image quality, specifically including the following steps:

1) constructing training data: obtaining a clear artifact-free CT image as a target image; carrying out spiral CT scanning on a phantom corresponding to a target image to obtain spiral scanning conical beam projection data, and carrying out image reconstruction on the spiral scanning conical beam projection data by using the conventional image reconstruction method to obtain an initial image; carrying out a numerical simulation process on the initial image to generate corresponding simulated cone beam projection data; reconstructing the simulated cone beam projection data by adopting the same image reconstruction method as the initial image to obtain a secondary image;

2) building a convolutional neural network shown in fig. 8, and adding one input channel to an input layer of the convolutional neural network, that is, the built convolutional neural network has two input channels; respectively sending the initial image and the secondary image into two input channels, extracting gray information and structural information of the initial image and the secondary image through a convolutional neural network, and estimating a real image after autonomous learning is carried out according to the extracted gray information and structural information;

3) constructing a loss function about a target image and an estimated real image, and training the convolutional neural network by using a gradient descent method; the loss function is:

wherein Img denotes an output image of the convolutional neural network, Img_tureRepresenting the target image corresponding to Img, Img_kThe value of the pixel representing the k-th pixel in the image Img, Img_ture,kRepresenting an image Img_tureThe representing pixel value of the kth pixel point;

4) respectively adopting the method in the step 1) to obtain an initial image and a secondary image for the low-voltage spiral scanning cone beam projection data and the high-voltage spiral scanning cone beam projection data obtained in the step 1); and sending the obtained initial image and the secondary image into a trained convolutional neural network to obtain a corresponding clear artifact-free high/voltage reconstructed image.

In the above improved spiral reconstruction method, the existing image reconstruction method includes, but is not limited to, a filtered back projection method, an iterative reconstruction method, and a compressive sensing method, and the convolutional neural network includes: CNN, ResNet and Unet.

Step 3: the reconstructed images of the two voltages obtained can be directly subjected to material decomposition in the image domain. In the material decomposition, the decomposition function F of the polynomial of the image domain can be solved by the following method:

wherein, F_wat、F_iodThe functions respectively representing the threshold material decomposition of the image to be solved are usually represented by polynomials; j represents an index value of each pixel of the inputted high voltage image/low voltage image; img_low、Img_highRespectively representing a low-voltage image and a high-voltage image, Img, obtained by image reconstruction_wat、Img_iodRepresenting the true water-based image and iodine-based image, respectively, can often be set up in advance by phantom measurements or numerical simulations.

The current commercial CT system capable of dual energy scanning by high voltage switching is the high speed switching spectrum CT of GE. The product needs to change the design of the high-voltage generator on the aspect of hardware, and simultaneously reduces the requirement on voltage stabilization to achieve the aim of high-speed switching. This implementation has a cost investment in hardware, the energy spectrum scan of the system requires special calibration, and the exposure current cannot be modified at will in the scan protocol, resulting in no current modulation to reduce the radiation dose.

Although the dual-energy scanning is realized by high-voltage switching as in the GE, the dual-energy scanning does not need to be switched every sampling as in the GE system. The switching speed can be adjusted according to the conditions of the existing hardware, and the hardware does not need to be changed at all, so the realization cost is extremely low. Due to the reduced switching speed, the voltage can maintain the same stability as a single energy scan, and therefore no special calibration is required. The exposure current is also easily modified and has the ability to achieve automatic current modulation. Generally, compared with the GE high-speed switching mode, the invention mainly has the advantages of low cost, easy realization, large energy separation and low dosage.

The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.