A high-resolution crystal imaging detector signal processing method and system

Abstract

This invention relates to a high-resolution crystal imaging detector signal processing method and system. The method involves acquiring multiple crystal signal waveforms at a high sampling rate; calculating integral values ​​from these waveforms to convert them into corresponding position signals; mapping these position signals onto the same image to obtain a crystal image; and performing multi-layer convolution and pooling operations on the crystal image using a neural network algorithm to extract region boundaries and structural information. The crystal image is then segmented at the pixel level to obtain a crystal array. The neural network algorithm is based on machine learning. Acquiring high-sampling-rate crystal signal waveforms to obtain a crystal image containing position information improves the integrity of the signal waveforms and effectively avoids information loss. Simultaneously, the use of a neural network algorithm for pixel-level segmentation of the crystal image enhances position resolution, reduces edge crystal adhesion, and improves the positional clarity of the crystal array.

Description

Technical Field

[0001] This invention relates to the field of nuclear radiation signal algorithm technology, and in particular to a high-resolution crystal imaging detector signal processing method and system. Background Technology

[0002] Positron emission tomography (PET) is a non-invasive medical imaging technique primarily used to observe metabolic and biochemical processes within the human body. Its physical basis is positron annihilation, where the decay of a radioactive isotope releases positrons that collide and annihilate with surrounding electrons, producing a pair of 511 keV gamma photons (crystals) pointing in opposite directions. PET detectors can determine the distribution of radioactive tracers within the body by capturing this pair of photons (crystal signals). Unlike structural imaging techniques such as CT and MRI, PET directly reflects functional information such as tissue metabolism, blood flow, or receptor density by tracking radiolabeled bioactive molecules, enabling the early detection of abnormalities in diseases. Its core components include radioactive tracers, a ring detector array, and image reconstruction algorithms.

[0003] Traditional PET detectors mostly use the peak method to acquire crystal signals, but due to limitations in sampling rate and signal processing methods, they suffer from low position resolution, and the captured edge crystal signals are prone to sticking together, resulting in a blurred position map. At the same time, the system sensitivity is limited, and the low sampling rate cannot completely capture the crystal signal waveform, resulting in serious information loss.

[0004] The above problems urgently need to be addressed. Summary of the Invention

[0005] This invention discloses a high-resolution crystal imaging detector signal processing method and system, which aims to solve the technical problems existing in the prior art.

[0006] The present invention adopts the following technical solution: On one hand, the present invention provides a high-resolution crystal imaging detector signal processing method, which includes: acquiring multiple crystal signal waveforms with high sampling rates; calculating integral values ​​of the multiple crystal signal waveforms and converting the multiple crystal signal waveforms into corresponding position signals respectively; mapping the multiple position signals onto the same image to obtain a crystal image; and performing multi-layer convolution and pooling operations on the crystal image based on a neural network algorithm to extract the region boundaries and structural information in the crystal image, and performing pixel-level segmentation on the crystal image to obtain a crystal array, wherein the neural network algorithm is based on machine learning.

[0007] Optionally, acquiring multiple crystal signal waveforms with a high sampling rate includes: using N analog-to-digital converters to operate in parallel, each analog-to-digital converter acquiring M crystal signal waveforms, where M and N are both positive integers greater than 1; stacking the M crystal signal waveforms alternately according to the acquisition time order to obtain A crystal signal waveforms; filtering the A crystal signal waveforms to filter out identical crystal signal waveforms at the same time point to obtain B crystal signal waveforms with a high sampling rate, where A and B are both positive integers greater than 1, and A is greater than or equal to B, and A is the product of M and N.

[0008] Optionally, the step of calculating the integral value of the plurality of crystal signal waveforms and converting the plurality of crystal signal waveforms into corresponding position signals includes: determining the integration interval based on the crystal signal waveforms by pulse triggering; performing integration calculation within the integration interval using the trapezoidal rule to determine the integral value, wherein the integral value corresponds to crystal energy information; and determining the crystal position signal by the energy distribution in the crystal energy information.

[0009] Optionally, before mapping multiple position signals onto the same image to obtain crystal imaging, the method further includes: convolving the position signals with a Gaussian kernel function to obtain filtered position signals; and mapping the filtered position signals onto a target interval to obtain normalized position signals.

[0010] Optionally, mapping multiple position signals onto the same image to obtain crystal imaging includes: setting the pixel size in the image according to the physical size of the crystal array, and establishing a mapping relationship between position coordinates and pixels; traversing multiple normalized position signals, calculating the pixel index corresponding to each of the multiple normalized position signals, accumulating the counts of the same pixel index to obtain a pixel count; constructing a pixel count matrix based on the pixel count; and converting the pixel count matrix into a grayscale image or a pseudo-color image to obtain crystal imaging, wherein the higher the pixel count, the brighter the pixel position is displayed.

[0011] Optionally, the step of performing multi-layer convolution and pooling operations on the crystal imaging based on the neural network algorithm to extract region boundaries and structural information in the crystal imaging, and performing pixel-level segmentation on the crystal imaging to obtain a crystal array, includes: taking the crystal imaging as input and inputting it to the input layer of the neural network algorithm; the input layer is connected to an encoder, the encoder including three sets of convolutional pooling structures, each set of convolutional pooling structures including two convolutional layers and one pooling layer, the two convolutional layers are followed by a pooling layer, the crystal imaging extracts pixel-level image features through the convolutional layers, and the pooling layer connected after the two convolutional layers processes the image output by the convolutional layers. Downsampling is performed to reduce the size of the output image. The crystal imaging is progressively processed through three convolution and pooling operations to obtain a feature map. The image features include edges, corners, and textures in the crystal imaging. The encoder is connected to the decoder, which includes three sets of deconvolution layers. The feature map is progressively upsampled through the three sets of deconvolution layers to restore the feature map to the same size as the crystal imaging, resulting in a crystal feature map with the same size as the crystal imaging and pixel-level image features. The decoder is connected to the output layer, which segments the crystal feature map according to pixel-level image features to obtain a crystal array.

[0012] Optionally, the input layer converts the crystal image into a single-channel grayscale image before inputting it into the encoder; the convolutional layer uses the ReLU activation function to enhance the nonlinear expression.

[0013] Optionally, the output layer uses the Sigmoid activation function to output the probability distribution of each pixel belonging to the crystal array region while segmenting according to pixel-level image features; the output layer uses 1×1 convolution to convert the crystal feature map into the same number of channels as the number of crystal arrays, wherein the number of channels is used to indicate the probability map of each crystal in the crystal array, and each channel corresponds to the spatial distribution probability of a crystal.

[0014] According to another aspect of the present invention, a high-resolution crystal imaging detector signal processing system is also provided, comprising: an acquisition module for acquiring multiple crystal signal waveforms at a high sampling rate; a position conversion module for calculating integral values ​​of the multiple crystal signal waveforms and converting the multiple crystal signal waveforms into corresponding position signals respectively; an imaging module for mapping the multiple position signals onto the same image to obtain a crystal image; and a network segmentation module for performing multi-layer convolution and pooling operations on the crystal image based on a neural network algorithm to extract region boundaries and structural information in the crystal image, and performing pixel-level segmentation on the crystal image to obtain a crystal array, wherein the neural network algorithm is based on machine learning.

[0015] According to another aspect of the present invention, a non-volatile storage medium is also provided, the non-volatile storage medium storing a plurality of instructions adapted for loading by a processor and executing any one of the high-resolution crystal imaging detector signal processing methods described herein.

[0016] The technical solution adopted in this invention can achieve at least one of the following beneficial effects: In this embodiment of the invention, multiple crystal signal waveforms with high sampling rates are acquired; integral values ​​are calculated on the multiple crystal signal waveforms to convert them into corresponding position signals; the multiple position signals are mapped onto the same image to obtain a crystal image; based on a neural network algorithm, the crystal image is subjected to multi-layer convolution and pooling operations to extract the region boundaries and structural information in the crystal image, and the crystal image is segmented at the pixel level to obtain a crystal array. The neural network algorithm is based on machine learning. This achieves the goal of acquiring crystal signal waveforms with high sampling rates to obtain crystal images containing position information, while simultaneously using a neural network algorithm to segment the crystal image at the pixel level. This improves the integrity of the signal waveform, effectively avoids information loss, enhances position resolution, effectively avoids edge crystal adhesion, and improves the positional clarity of the crystal array. Furthermore, it solves the technical problems of traditional PET detectors, which often use peak sampling to acquire crystal signals, but are limited by sampling rate and signal processing methods, resulting in low position resolution, easy adhesion of captured edge crystal signals leading to blurred position maps, and limited system sensitivity, where low sampling rates cannot completely capture crystal signal waveforms, leading to severe information loss. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below, forming part of the present invention. The illustrative embodiments of the present invention and their descriptions explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings: Figure 1 This is a flowchart of a high-resolution crystal imaging detector signal processing method according to Embodiment 1 of the present invention; Figure 2 This is an actual crystal array diagram in a high-resolution crystal imaging detector signal processing method according to Embodiment 1 of the present invention; Figure 3 This is a crystal array diagram segmented based on an algorithm in a high-resolution crystal imaging detector signal processing method according to Embodiment 1 of the present invention; Figure 4 This is a model structure diagram of the neural network model in a high-resolution crystal imaging detector signal processing method according to Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the structure of a high-resolution crystal imaging detector signal processing system according to Embodiment 2 of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. In the description of this invention, it should be noted that the term "or" is generally used to include the meaning of "and / or," unless otherwise expressly indicated.

[0019] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or a magnetic connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. Furthermore, in the description of this application, the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. In the description of this invention, "a plurality of" means at least two, such as two, three, or more, unless otherwise explicitly specified.

[0020] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0021] To address the problems existing in related technologies, this application provides a high-resolution crystal imaging detector signal processing method and system.

[0022] Example 1 This embodiment provides a high-resolution crystal imaging detector signal processing method, such as... Figure 1 As shown, Figure 1 This is a flowchart of a high-resolution crystal imaging detector signal processing method according to Embodiment 1 of the present invention, the method comprising: Step S102: Acquire multiple crystal signal waveforms with high sampling rate; In some preferred embodiments, acquiring multiple crystal signal waveforms with a high sampling rate includes: using N analog-to-digital converters (ADCs) to operate in parallel, with each ADC acquiring M crystal signal waveforms, where M and N are both positive integers greater than 1; stacking the M crystal signal waveforms alternately according to the acquisition time order to obtain A crystal signal waveforms; filtering the A crystal signal waveforms to remove identical crystal signal waveforms at the same time point to obtain B crystal signal waveforms with a high sampling rate, where A and B are both positive integers greater than 1, and A is greater than or equal to B, and A is the product of M and N.

[0023] Optionally, to improve the completeness of the signal waveform and effectively avoid information loss, it is necessary to increase the signal sampling rate to obtain more crystal signal waveforms, thus obtaining a complete crystal signal. There are two specific methods: One is to replace the hardware, which has a direct and significant effect by using a higher-performance analog-to-digital converter (ADC), i.e., a high-speed ADC chip. In this case, a high-performance ADC is used to more accurately obtain the signal characteristics, such as shape, amplitude, and timing performance. After determining the signal waveform characteristics, an algorithm is used, and later experiments can convert the signal waveform characteristics into a crystal signal waveform using a lower-cost, lower-performance ADC. The other method is interleaved sampling (an indirect method), using multiple low-speed ADCs working in parallel and then interleaving their output data to simulate a high-speed ADC.

[0024] Step S104: Calculate the integral value of multiple crystal signal waveforms and convert the multiple crystal signal waveforms into corresponding position signals respectively; Optionally, the integral value can be calculated based on the detector characteristics. The crystal signal waveform is proportional to the energy, and the crystal signal waveform is a function of voltage over time. Where Q is the peak height of the crystal signal waveform, and i is the i-th acquisition. Let i be the voltage value collected in the i-th sampling. The time interval between the i-th data collection and the (i-1)-th data collection is denoted as .

[0025] Optionally, the signal-to-noise ratio can be improved by calculating the integral value. Integration is a low-pass filtering process that accumulates the changing signal and averages out rapidly changing high-frequency random noise, causing positive and negative cancellation, thereby suppressing high-frequency noise components. By knowing the high-frequency noise components in the crystal signal waveform before converting the position signal, the obtained position signal becomes more accurate.

[0026] In some preferred embodiments, the integration values ​​of multiple crystal signal waveforms are calculated, and the multiple crystal signal waveforms are converted into corresponding position signals, including: determining the integration interval based on the crystal signal waveforms by pulse triggering; performing integration calculations using the trapezoidal rule within the integration interval to determine the integration value, wherein the integration value corresponds to crystal energy information; and determining the crystal position signal by the energy distribution in the crystal energy information.

[0027] Optionally, the crystal energy information is directly proportional to the integral value. The integral value can be used to obtain the crystal energy information. After obtaining the crystal energy information, the position signal of the crystal can be determined based on the crystal energy information. In the position-sensitive detector, the voltage amplitude values ​​of multiple output terminals of the signal are directly related to the spatial position of the particle incident. This is related to the circuit design, such as the Angel circuit.

[0028] Specifically, since the integral value is not entirely energy information, but is a conversion of the crystal signal waveform acquired and output by the PET detector circuit, the integral value is also output in the PET detector circuit. The output is converted from the information carried in the charge of the hardware circuit. The PET detector circuit uses an Angel circuit, which can use a resistor chain to divide and read out the charge, thereby forming a crystal signal waveform, which is then converted into an integral value.

[0029] Optionally, the charge signal on the anode can be divided into multiple readout channels via a resistor chain. In PET detectors, each PMT or SiPM in the array itself acts as a charge divider, "splitting" the energy of the scintillation photons (in the form of an integral value) to different spatial locations. The signal is processed using a charge-sensitive preamplifier and a shaping amplifier, and the analog signal is converted to a digital signal using an analog-to-digital converter (ADC). The PET detector readout circuitry employs a similar approach, amplifying, shaping, and digitizing the signal from the SiPM or PMT via front-end electronics.

[0030] Optionally, the X and Y coordinates of the interaction can be calculated in real time using a series of analog addition and division circuits. Assuming a 2x2 PMT array for positioning, with indices A, B, C, and D, and the signal amplitude output by each PMT denoted as S... A S B S C S D It is proportional to the energy of the deposited crystal.

[0031] First, calculate the total energy Z: sum the signals of all PMTs to obtain the total signal Z as follows: The Z signal is proportional to the total energy of gamma photons deposited in the crystal.

[0032] Next, calculate the X and Y position signals. Specifically, the X position is calculated by adding the PMT signals in the positive X-axis direction and subtracting the sum of the PMT signals in the negative X-axis direction. This difference is then divided by the total energy Z for normalization. The Y position is calculated in the same way as the X position. The specific X and Y position signals are as follows: Optionally, in a PET detector, a digital signal processor or field-programmable gate array (FPGA) can be used to execute the above charge centroid formula, thereby quickly and in real time calculating the X and Y positions of the scintillation event to obtain the precise position of the crystal.

[0033] Step S106: Map multiple position signals onto the same image to obtain crystal imaging; In some preferred embodiments, before mapping multiple position signals onto the same image to obtain crystal imaging, the method further includes: convolving the position signals with a Gaussian kernel function to obtain filtered position signals; and mapping the filtered position signals onto a target interval to obtain normalized position signals.

[0034] Optionally, a filter can be used to remove high-frequency noise from the position signal, and normalization can be performed simultaneously. During normalization, the filtered position signal is mapped to a fixed numerical range, such as [0, 255], to meet the input requirements of image processing. A crystal image (e.g., ...) is obtained from the position signal. Figure 3 ).

[0035] In some preferred embodiments, mapping multiple position signals onto the same image to obtain crystal imaging includes: setting the pixel size in the image according to the physical size of the crystal array, and establishing a mapping relationship between position coordinates and pixels; traversing multiple normalized position signals, calculating the pixel index corresponding to each of the multiple normalized position signals, and accumulating the counts of the same pixel index to obtain a pixel count; constructing a pixel count matrix based on the pixel count; and converting the pixel count matrix into a grayscale image or a pseudo-color image to obtain crystal imaging, wherein pixels with higher pixel counts are displayed brighter.

[0036] Optionally, a mapping relationship is first established. Based on the actual physical size of the crystal array, the size of each pixel in the image is determined, thereby establishing the correspondence between the crystal position coordinates and the image pixels, laying the foundation for subsequent signal localization.

[0037] Next, the pixel count is calculated and accumulated. Each normalized position signal is processed one by one, and their corresponding pixel index is calculated. Then, the counts of signals with the same pixel index are accumulated to obtain the cumulative count for each pixel. This step can reflect the density of signals at different positions.

[0038] Reconstruct the pixel counting matrix. Based on the counts of each pixel obtained in the previous step, construct a pixel counting matrix. The value of each element in the matrix corresponds to the cumulative count of the corresponding pixel.

[0039] Finally, the pixel count matrix is ​​converted into a visual image, either a grayscale or pseudocolor image. Since the higher the pixel count, the brighter the image appears, the distribution of signals in the crystal can be visually presented through the brightness (or color depth) of the image, ultimately yielding the crystal imaging result.

[0040] The entire process establishes a correlation between physical location signals and image pixels, and through counting accumulation and visualization transformation, it achieves an intuitive presentation of the crystal signal distribution.

[0041] Step S108: Based on the neural network algorithm, multi-layer convolution and pooling operations are performed on the crystal imaging to extract the region boundary and structural information in the crystal imaging, and the crystal imaging is segmented at the pixel level to obtain the crystal array. The neural network algorithm is based on machine learning.

[0042] Optionally, specifically, we first construct a neural network model, which is as follows: Input layer: Receives pre-processed crystal images.

[0043] Convolutional layers: The neural network extracts pixel locations from the input crystal image and performs sliding operations and weighted summations on local regions of the crystal image using different convolutional kernels to obtain feature maps. These feature maps represent information such as edges, corners, and textures in the crystal image.

[0044] Pooling layer: After convolution, downsampling is used to reduce the size of the feature map, reduce the amount of computation, and enhance the translation invariance of the model to obtain compressed, high-level features.

[0045] Deconvolutional layers: For segmentation tasks, compressed feature maps need to be enlarged back to their original image size for pixel-level classification. This is achieved through deconvolution and upsampling, combined with feature maps from the encoder to recover spatial details.

[0046] Output layer: It is a 1×1 convolutional layer followed by an activation function, which outputs a segmentation probability map of the same size as the input image, where the value of each pixel represents the probability of it belonging to "which crystal".

[0047] Specifically, neural network models also include loss functions and optimizers. The loss function can characterize the difference between the model's prediction and the true value, while the optimizer can calculate the gradient of the loss function with respect to all weights and biases in the neural network through backpropagation.

[0048] Finally, the neural network model needs to be trained. Using optimizers and gradients, the network parameters are iteratively updated to minimize the loss function until the model performance converges.

[0049] Optionally, the constructed neural network model can be applied to actual images, as follows: Input crystal imaging: The generated crystal image to be segmented is input into the trained neural network model.

[0050] Forward propagation and feature extraction: The input crystal image is propagated forward using the same multi-layer convolution and pooling operations as in the training phase; the neural network model identifies and extracts high-level features in the crystal image, such as the shape, boundaries, and internal brightness uniformity of the crystal.

[0051] Segmentation output: In the decoder and output layers, the model classifies each pixel in the crystal image based on the extracted features; according to the brightness distribution and geometric features in the crystal image, the brightness pixels at multiple locations are segmented to obtain the initial segmentation.

[0052] Further post-processing can be performed on the initial segmentation results to remove excessively small noise points and smooth edges. Additionally, after obtaining the final segmentation mask, quantitative features of the crystal can be further extracted, such as the average brightness, brightness standard deviation, and uniformity within the crystal. This ultimately yields the optimal crystal segmentation result image (crystal array), i.e., the precise crystal outline, and a list of extracted crystal feature parameters, such as... Figure 2 and Figure 3 As shown, Figure 2 This is an actual crystal array diagram in a high-resolution crystal imaging detector signal processing method according to Embodiment 1 of the present invention. Figure 3 This is a crystal array diagram segmented based on an algorithm in a high-resolution crystal imaging detector signal processing method according to Embodiment 1 of the present invention.

[0053] In some preferred embodiments, based on a neural network algorithm, multi-layer convolution and pooling operations are performed on the crystal imaging to extract region boundaries and structural information from the crystal imaging, and pixel-level segmentation is performed on the crystal imaging to obtain a crystal array. This includes: taking the crystal imaging as input and feeding it into the input layer of the neural network algorithm; the input layer is connected to an encoder, which includes three sets of convolutional pooling structures, each set including two convolutional layers and one pooling layer, with the pooling layer following the two convolutional layers; the crystal imaging extracts pixel-level image features through the convolutional layers, and the pooling layer following the two convolutional layers performs convolutional pooling on the crystal imaging. The output image is downsampled to reduce its size. The crystal imaging is then subjected to three convolution and pooling operations to obtain a feature map. The image features include edges, corners, and textures in the crystal imaging. The encoder connects to the decoder, which consists of three sets of deconvolution layers. The feature map is then upsampled through these three sets of deconvolution layers to restore it to the same size as the crystal imaging, resulting in a crystal feature map with pixel-level image features. The decoder connects to the output layer, which segments the crystal feature map according to pixel-level image features to obtain a crystal array.

[0054] Optionally, neural network segmentation and feature extraction for crystal imaging includes everything from data preparation, model building and training, to the final application and result output.

[0055] The first step is the preparatory work before building and training the neural network, which involves learning to accurately segment individual crystals from a large amount of crystal imaging data.

[0056] The input data consists of crystal images stored in matrix form, where the count value of each pixel represents its brightness. Data preparation involves manually labeling a large number of crystal image samples to create a "ground truth" segmentation map for each image. This is a binary mask where crystal regions are represented by 1s and inter-crystal boundaries by 0s. Preprocessing involves uniformly processing all crystal images to ensure the standardization of the input data. Size normalization involves scaling all crystal images to the same size, 256×256 pixels. Normalization scales pixel values ​​from their original range to between [0,1], accelerating model convergence.

[0057] Optionally, after the preparation work is completed, a neural network model needs to be built. This neural network model uses the U-Net structure as its basis, combining downsampling and upsampling paths to simultaneously capture global contextual information and local fine details. For example... Figure 4 As shown, Figure 4 The following is a model structure diagram of the neural network model in a high-resolution crystal imaging detector signal processing method according to Embodiment 1 of the present invention: Input layer: Receives a pre-processed 256×256 pixel grayscale image.

[0058] The encoder consists of multiple convolutional and pooling layers: it comprises multiple downsampling blocks. Each downsampling block contains two consecutive convolutional layers followed by a pooling layer. Convolutional layers: 3×3 kernels are used with a stride of 1, and "same" padding is used to maintain the feature map size. Each convolutional layer is followed by an activation function. Activation function: ReLU is chosen, with the formula f(x) = max(0,x). This effectively solves the vanishing gradient problem and accelerates model training. Pooling layers: 2×2 max pooling is used with a stride of 2. This halves the feature map size while preserving the most important feature information, enhancing the model's translation invariance.

[0059] Decoder: Upsampling is performed using deconvolution with a 2×2 convolution kernel and a stride of 2, which doubles the size of the feature map.

[0060] Skip connections: After each upsampling step in the decoder, its result is concatenated with the feature map of the corresponding layer in the encoder. This allows the model to fully utilize the high-resolution spatial details captured in the encoder when reconstructing the image size, resulting in more accurate segmentation boundaries.

[0061] Output layer: A 1×1 convolutional layer used to reduce the number of channels to 1.

[0062] The Sigmoid activation function is , The Sigmoid function maps the output value of each pixel to the range [0,1], representing the probability that the pixel belongs to the crystal region.

[0063] Specifically, this also includes: loss function and optimizer Loss function: The binary cross-entropy loss function is selected, and its formula is as follows: Where y is the true value. It represents the probability predicted by the model. It is a good measure of the model's accuracy in classifying each pixel as either a crystal or the background.

[0064] Optimizer: The Adam optimizer is selected. Adam is an adaptive learning rate optimization algorithm with fast convergence speed.

[0065] In some preferred embodiments, the input layer converts the crystal image into a single-channel grayscale image before inputting it to the encoder; the convolutional layer uses the ReLU activation function to enhance the nonlinear expression.

[0066] In some preferred embodiments, the output layer uses the Sigmoid activation function to output the probability distribution of each pixel belonging to the crystal array region while segmenting according to pixel-level image features; the output layer uses 1×1 convolution to convert the crystal feature map into the same number of channels as the number of crystal arrays, wherein the number of channels is used to indicate the probability map of each crystal in the crystal array, and each channel corresponds to the spatial distribution probability of a crystal.

[0067] Through the above steps S102 to S108, the goal of obtaining a crystal signal waveform with a high sampling rate to obtain a crystal image containing position information is achieved. At the same time, a neural network algorithm is used to perform pixel-level segmentation of the crystal image, thereby improving the integrity of the signal waveform, effectively avoiding information loss, enhancing position resolution, effectively avoiding edge crystal adhesion, and improving the position clarity of the crystal array. This solves the technical problems that traditional PET detectors mostly use the peak method to acquire crystal signals, but are limited by the sampling rate and signal processing method, resulting in low position resolution, easy adhesion of captured edge crystal signals, and blurred position map; at the same time, the system sensitivity is limited, and the low sampling rate cannot completely capture the crystal signal waveform, resulting in serious information loss.

[0068] Example 2 This embodiment also provides a high-resolution crystal imaging detector signal processing system for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the terms "module" and "system" can refer to a combination of software and / or hardware that performs a predetermined function. Although the systems described in the following embodiments are preferably implemented in software, hardware implementations, or combinations of software and hardware, are also possible and contemplated.

[0069] According to embodiments of the present invention, a system embodiment for implementing the above-described high-resolution crystal imaging detector signal processing method is also provided. Figure 5 This is a schematic diagram of the structure of a high-resolution crystal imaging detector signal processing system according to Embodiment 2 of the present invention, as shown below. Figure 5 As shown, the above system includes: an acquisition module 201, a position conversion module 202, an imaging module 203, and a network segmentation module 204, wherein: Acquisition module 201 is used to acquire multiple crystal signal waveforms at a high sampling rate; The position conversion module 202 is connected to the acquisition module 201 and is used to calculate the integral value of multiple crystal signal waveforms and convert the multiple crystal signal waveforms into corresponding position signals respectively. Imaging module 203, connected to position conversion module 202, is used to map multiple position signals onto the same image to obtain crystal imaging; The network segmentation module 204, connected to the imaging module 203, is used to perform multi-layer convolution and pooling operations on the crystal imaging based on the neural network algorithm, extract the region boundary and structural information in the crystal imaging, and perform pixel-level segmentation on the crystal imaging to obtain the crystal array. The neural network algorithm is based on machine learning.

[0070] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0071] It should be noted that the acquisition module 201, position conversion module 202, imaging module 203, and network segmentation module 204 mentioned above correspond to steps S102 to S108 in the embodiments. The instances and application scenarios implemented by the above modules and their corresponding steps are the same, but they are not limited to the content disclosed in the above embodiments. It should be noted that the above modules can run on a computer terminal as part of the system.

[0072] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0073] The aforementioned high-resolution crystal imaging detector signal processing system may further include a processor and a memory. The aforementioned acquisition module 201, position conversion module 202, imaging module 203, and network segmentation module 204 are all stored in the memory as program modules, and the processor executes the aforementioned program modules stored in the memory to realize the corresponding functions.

[0074] The processor contains a core that retrieves the corresponding program modules from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0075] According to an embodiment of this application, an embodiment of a non-volatile storage medium is also provided. Optionally, in this embodiment, the non-volatile storage medium includes a stored program, wherein, when the program is running, it controls the device where the non-volatile storage medium is located to execute any of the aforementioned high-resolution crystal imaging detector signal processing methods.

[0076] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals, and the non-volatile storage medium includes stored programs.

[0077] Optionally, during program execution, the device containing the non-volatile storage medium can be controlled to perform the following functions: acquire multiple crystal signal waveforms with high sampling rates; calculate the integral value of the multiple crystal signal waveforms and convert them into corresponding position signals respectively; map the multiple position signals onto the same image to obtain crystal imaging; and perform multi-layer convolution and pooling operations on the crystal imaging based on a neural network algorithm to extract the region boundaries and structural information in the crystal imaging, and perform pixel-level segmentation on the crystal imaging to obtain a crystal array, wherein the neural network algorithm is based on machine learning.

[0078] According to an embodiment of this application, an embodiment of a processor is also provided. Optionally, in this embodiment, the processor is used to run a program, wherein the program executes any of the above-described high-resolution crystal imaging detector signal processing methods.

[0079] According to an embodiment of this application, an embodiment of a computer program product is also provided. Optionally, in this embodiment, the computer program product includes a computer program that, when executed by a processor, implements the steps of any of the above-described high-resolution crystal imaging detector signal processing methods.

[0080] Optionally, when the aforementioned computer program product is executed on a data processing device, it is suitable to execute an initialization program with the following method steps: acquiring multiple crystal signal waveforms with a high sampling rate; calculating the integral value of the multiple crystal signal waveforms and converting the multiple crystal signal waveforms into corresponding position signals respectively; mapping the multiple position signals onto the same image to obtain a crystal image; and performing multi-layer convolution and pooling operations on the crystal image based on a neural network algorithm to extract the region boundary and structural information in the crystal image, and performing pixel-level segmentation on the crystal image to obtain a crystal array, wherein the neural network algorithm is based on machine learning.

[0081] This invention provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: acquiring multiple crystal signal waveforms with a high sampling rate; calculating the integral value of the multiple crystal signal waveforms and converting them into corresponding position signals; mapping the multiple position signals onto the same image to obtain a crystal image; and performing multi-layer convolution and pooling operations on the crystal image based on a neural network algorithm to extract the region boundaries and structural information in the crystal image, and performing pixel-level segmentation on the crystal image to obtain a crystal array. The neural network algorithm is based on machine learning.

[0082] The order of the above embodiments of the present invention is merely for description and does not represent the superiority or inferiority of the embodiments.

[0083] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0084] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The system embodiments described above are merely illustrative; for example, the division of modules described above can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between modules, and may be electrical or other forms.

[0085] The modules described above as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0086] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0087] If the aforementioned integrated modules are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable non-volatile storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a non-volatile storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned non-volatile storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0088] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A signal processing method for a high-resolution crystal imaging detector, characterized in that, include: Acquire multiple crystal signal waveforms at high sampling rates; The multiple crystal signal waveforms are integrated to calculate the integral value, and the multiple crystal signal waveforms are converted into corresponding position signals respectively; By mapping multiple positional signals onto the same image, crystal imaging is obtained; Based on a neural network algorithm, the crystal imaging is subjected to multi-layer convolution and pooling operations to extract the region boundaries and structural information in the crystal imaging, and the crystal imaging is segmented at the pixel level to obtain a crystal array. The neural network algorithm is based on machine learning. The crystal image is used as input to the input layer of a neural network algorithm. The input layer is connected to an encoder, which includes three sets of convolutional pooling structures. Each set of convolutional pooling structures includes two convolutional layers and one pooling layer. The two convolutional layers are followed by a pooling layer. The crystal image extracts pixel-level image features through the convolutional layers. The pooling layer connected after the two convolutional layers downsamples the image output by the convolutional layers to reduce the size of the output image. The crystal image undergoes three convolutional and pooling operations to obtain a feature map. The image features include edges, corners, and textures in the crystal image. The encoder is connected to a decoder, which includes three sets of deconvolutional layers. The feature map is upsampled through the three sets of deconvolutional layers to restore the feature map to the same size as the crystal image, resulting in a crystal feature map with the same size as the crystal image and pixel-level image features. The decoder is connected to an output layer, which segments the crystal feature map according to pixel-level image features to obtain a crystal array. The output layer uses the Sigmoid activation function to output the probability distribution of each pixel belonging to the crystal array region while segmenting according to pixel-level image features.

2. The high-resolution crystal imaging detector signal processing method according to claim 1, characterized in that, The acquisition of multiple crystal signal waveforms at a high sampling rate includes: N analog-to-digital converters are used to work in parallel. Each analog-to-digital converter acquires M crystal signal waveforms, where M and N are both positive integers greater than 1. The M crystal signal waveforms are stacked alternately according to the acquisition time sequence to obtain A crystal signal waveforms; The A crystal signal waveforms are filtered to remove identical crystal signal waveforms at the same time point, resulting in B crystal signal waveforms with a high sampling rate. Here, A and B are both positive integers greater than 1, and A is greater than or equal to B. A is the product of M and N.

3. The high-resolution crystal imaging detector signal processing method according to claim 1, characterized in that, The step of integrating the multiple crystal signal waveforms and converting them into corresponding position signals includes: Based on the crystal signal waveform, the integration interval is determined by pulse triggering; Within the integration interval, the trapezoidal rule is used to perform integration calculations and determine the integral value, wherein the integral value corresponds to the crystal energy information; The position signal of the crystal is determined by the distribution of energy in the crystal energy information.

4. The high-resolution crystal imaging detector signal processing method according to claim 1, characterized in that, Before mapping multiple location signals onto the same image to obtain a crystal image, the method further includes: The position signal is convolved by a Gaussian kernel function to obtain a filtered position signal; The filtered position signal is mapped to the target interval to obtain a normalized position signal.

5. The high-resolution crystal imaging detector signal processing method according to claim 4, characterized in that, The process of mapping multiple position signals onto the same image to obtain crystal imaging includes: Based on the physical dimensions of the crystal array, the pixel size in the image is set, and the mapping relationship between position coordinates and pixels is established; Traverse multiple normalized position signals, calculate the pixel index corresponding to each of the multiple normalized position signals, and accumulate the counts of the same pixel index to obtain the pixel count; Based on the pixel count, construct a pixel count matrix; The pixel count matrix is ​​converted into a grayscale image or a pseudo-color image to obtain a crystal image, wherein the pixel with the higher pixel count is displayed brighter.

6. The high-resolution crystal imaging detector signal processing method according to claim 1, characterized in that, The input layer converts the crystal image into a single-channel grayscale image, which is then input to the encoder. The convolutional layer uses the ReLU activation function to enhance the nonlinear expression.

7. The high-resolution crystal imaging detector signal processing method according to claim 1, characterized in that, The output layer uses 1×1 convolution to convert the crystal feature map into the same number of channels as the number of crystal arrays. The number of channels is used to indicate the probability map of each crystal in the crystal array, and each channel corresponds to the spatial distribution probability of a crystal.

8. A high-resolution crystal imaging detector signal processing system, characterized in that, include: The acquisition module is used to acquire multiple crystal signal waveforms at a high sampling rate; The position conversion module is used to calculate the integral value of the plurality of crystal signal waveforms and convert the plurality of crystal signal waveforms into corresponding position signals respectively; The imaging module is used to map multiple position signals onto the same image to obtain a crystal image; The network segmentation module is used to perform multi-layer convolution and pooling operations on the crystal image based on a neural network algorithm, extract the region boundaries and structural information in the crystal image, and perform pixel-level segmentation on the crystal image to obtain a crystal array. The neural network algorithm is based on machine learning. The network segmentation module is used to input the crystal imaging as input to the input layer of the neural network algorithm. The input layer is connected to an encoder, which includes three sets of convolutional pooling structures. Each set of convolutional pooling structures includes two convolutional layers and one pooling layer. The two convolutional layers are followed by a pooling layer. The crystal imaging extracts pixel-level image features through the convolutional layers. The pooling layer connected after the two convolutional layers downsamples the image output by the convolutional layers to reduce the size of the output image. The crystal imaging undergoes three convolutional and pooling operations to obtain a feature map, wherein the image features include edges, corners, and textures in the crystal imaging. The encoder is connected to a decoder, which includes three sets of deconvolutional layers. The feature map is upsampled through the three sets of deconvolutional layers to restore the feature map to the same size as the crystal imaging, resulting in a crystal feature map with the same size as the crystal imaging and pixel-level image features. The decoder is connected to an output layer, which segments the crystal feature map according to pixel-level image features to obtain a crystal array. The output layer uses a sigmoid function. The activation function, while segmenting according to pixel-level image features, outputs the probability distribution of each pixel belonging to the crystal array region.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores multiple instructions adapted for loading and execution by a processor of a high-resolution crystal imaging detector signal processing method according to any one of claims 1 to 7.

Images