Medical injury CT image processing method and system

Abstract

The invention provides a medical injury CT image processing method and system, and relates to the field of medical image processing, and the method comprises the steps: carrying out the data preprocessing of a CT image; determining the noise type of the noise in the CT image; determining the noise intensity of the noise in the CT image; determining a training data set according to the noise type and the noise intensity; according to the training data set, training the noise processing model to obtain a trained noise processing model; processing the CT image according to the trained noise processing model to obtain a denoised CT image; determining a damage area at a damage position in the CT image after noise reduction, and determining a second CT value of a pixel at the damage position; determining a damage severity coefficient according to the damage area and the second CT value; and generating a damage report according to the damage severity coefficient. According to the invention, the accuracy and effectiveness of detail analysis of the damage position in the CT image can be improved.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to a method and system for processing medical injury CT images. Background Art

[0002] In the related art, CN117237255A discloses a CT image processing method, device, storage medium and CT machine, which belong to the field of medical image processing, including: obtaining a first resolution CT image; inputting the first resolution CT image into a CT image processing model to obtain a second resolution CT image, and the resolution of the second resolution CT image is higher than the resolution of the first resolution CT image; wherein the CT image processing model is trained using a sample CT image and a labeled CT image with a resolution higher than the sample CT image, and the sample CT image is reconstructed based on projection data; this scheme can improve the resolution of CT images.

[0003] CN103971387A discloses a method for CT image reconstruction, including the following steps: obtaining original data according to CT scanning; correcting the original data to obtain corrected data; performing denoising on the corrected data to obtain denoised data; rearranging the denoised data to obtain rearranged data; convolving the rearranged data to obtain convolution data; performing back-projection processing on the convolution data to obtain a full scan field of view image and a clinically selected field of view image; obtaining an error image according to the full scan field of view image; and obtaining a CT image according to the clinically selected field of view image and the error image. The method of this scheme obtains an error image through a full scan field of view image, and obtains a CT image according to the clinically selected field of view image and the error image, which can ensure the resolution of the reconstructed CT image, and can greatly reduce the amount of calculation in the reconstruction process, and improve the speed of image reconstruction.

[0004] Based on the above related technologies, the resolution of CT images can be improved. However, the related technologies do not analyze the content in the CT images, that is, they cannot analyze the degree of damage to the damaged parts in the CT images.

[0005] The information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0006] The present invention provides a medical injury CT image processing method and system, which can solve the technical problem that the related technology cannot analyze the injury degree of the injury part in the CT image.

[0007] According to a first aspect of the present invention, a method for processing medical injury CT images is provided, comprising: Perform data preprocessing on CT images; Determine the noise type of noise in CT images; Determining the noise intensity of noise in CT images; Determining a training data set according to the noise type and the noise intensity, wherein the training noise type of the training CT image in the training data set is the same as the noise type; According to the training data set, the noise processing model is trained to obtain a trained noise processing model; Processing the CT image according to the trained noise processing model to obtain a denoised CT image; Determine the lesion area at the lesion location in the noise-reduced CT image, and determine a second CT value of the pixel at the lesion location; determining an injury severity coefficient according to the injury area and the second CT value; An injury report is generated based on the injury severity coefficient.

[0008] According to a second aspect of the present invention, a medical injury CT image processing system is provided, comprising: A preprocessing module, used for performing data preprocessing on CT images; A noise type module, used to determine the noise type of noise in the CT image; A noise intensity module for determining the noise intensity of noise in the CT image; A training data module, used to determine a training data set according to the noise type and the noise intensity, wherein the training noise type of the training CT image in the training data set is the same as the noise type; A training model module, used to train the noise processing model according to the training data set to obtain a trained noise processing model; A training model module, used to process the CT image according to the trained noise processing model to obtain a denoised CT image; An image analysis module, used to determine the damage area at the damage position in the CT image after noise reduction, and determine the second CT value of the pixel at the damage position; An injury severity module, used to determine an injury severity coefficient according to the injury area and the second CT value; The damage report module is used to generate a damage report according to the damage severity coefficient.

[0009] Technical effect: According to the present invention, the noise type and noise intensity in medical injury CT images can be accurately analyzed, and a training data set can be determined according to the noise type and noise intensity to train the noise processing model, and the injury area and CT value of the injury site can be determined according to the denoised medical injury CT images. Furthermore, the degree of injury at the injury location can be evaluated according to the injury area and CT value, which is conducive to improving the accuracy and effectiveness of the detailed analysis of the injury location in the CT image. When determining the noise intensity, the noise intensity of the noise in the CT image can be determined according to the CT value of the pixel and the number of pixels, which provides a data basis for the subsequent determination of the training data set and improves the calculation efficiency. When determining the peak signal-to-noise ratio, the peak signal-to-noise ratio can be determined according to the pixel value and the reference pixel value, which improves the convenience of calculation. When determining the training loss function, the influence of the spatial resolution, contrast resolution and peak signal-to-noise ratio on the noise intensity can be used to determine the influence of the above data on the error of predicting the noise intensity, so as to set the training loss function based on the influence, the relative error of predicting the noise intensity and the data volume of the training data set, so that the noise processing model can reduce the training loss function during the training process, and make the trained noise processing model more suitable for the noise reduction processing of the current type and intensity of CT images, and improve the accuracy of the noise processing model in a more targeted manner. When determining the injury severity coefficient, the injury severity coefficient is determined according to the second CT value of the first pixel sampling point, the second CT value of the second pixel sampling point and the injury area. In the calculation process, the most serious overall injury degree of the peripheral area in the injury position and the central area in the injury position can be determined according to the injury degree at the pixel sampling point in the similar shape. Further, the overall injury degree of the injury position is determined according to the overall injury degree of the peripheral area in the injury position and the central area in the injury position, and the comprehensive injury severity of the injury position is determined in combination with the injury area, thereby improving the comprehensiveness and accuracy of the injury severity coefficient.

[0010] It should be understood that the above general description and the following detailed description are exemplary and explanatory only and do not limit the present invention. Other features and aspects of the present invention will become more apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other embodiments can be obtained based on these drawings without creative work. Figure 1 A schematic diagram of a process flow of a medical injury CT image processing method according to an embodiment of the present invention is exemplarily shown; Figure 2 A block diagram of a medical injury CT image processing system according to an embodiment of the present invention is exemplarily shown. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0014] Figure 1 A schematic flow chart of a medical injury CT image processing method according to an embodiment of the present invention is exemplarily shown, and the method comprises: Step S101, performing data preprocessing on the CT image; Step S102, determining the noise type of the noise in the CT image; Step S103, determining the noise intensity of the noise in the CT image; Step S104, determining a training data set according to the noise type and the noise intensity, wherein the training noise type of the training CT image in the training data set is the same as the noise type; Step S105, training the noise processing model according to the training data set to obtain a trained noise processing model; Step S106, processing the CT image according to the trained noise processing model to obtain a denoised CT image; Step S107, determining the damage area at the damage position in the CT image after noise reduction, and determining the second CT value of the pixel at the damage position; Step S108, determining an injury severity coefficient according to the injury area and the second CT value; Step S109: Generate a damage report according to the damage severity coefficient.

[0015] According to the medical injury CT image processing method of the embodiment of the present invention, the noise type and noise intensity in the medical injury CT image can be accurately analyzed, and a training data set is determined according to the noise type and noise intensity to train the noise processing model, and the damage area and CT value of the damaged part are determined according to the medical injury CT image after noise reduction. Furthermore, the degree of damage at the damaged position is evaluated according to the damage area and CT value, which is beneficial to improving the accuracy and effectiveness of the detailed analysis of the damage position in the CT image.

[0016] According to an embodiment of the present invention, in step S101, data preprocessing is performed on the CT image.

[0017] For example, preprocessing operations such as normalization, cropping and enhancement are performed on CT images.

[0018] According to an embodiment of the present invention, in step S102, the noise type of the noise in the CT image is determined.

[0019] For example, by observing the distribution of noise in the image and determining the noise type of noise in the CT image, quantum noise usually appears as a randomly distributed granular appearance, while hardware system noise may appear as a more uniform or more concentrated noise pattern. Therefore, the noise type of noise in the CT image can be determined based on the characteristics of quantum noise and hardware system noise.

[0020] According to an embodiment of the present invention, in step S103, the noise intensity of the noise in the CT image is determined.

[0021] According to one embodiment of the present invention, step S103 includes: In the CT image, a preset area is determined; Obtaining the CT value and number of pixels of each pixel in a preset area; The noise intensity of the noise in the CT image is determined according to the CT value of the pixel and the number of the pixels.

[0022] For example, in a CT image, a preset area is selected. The preset area is located in a non-damaged area within the area where the organ is located in the CT image. The preset area should be as large and uniform as possible to accurately reflect the intensity of the noise. The CT value of each pixel in the preset area is obtained through the automatic calculation and display function of the CT scanner, and the number of pixels in the preset area is determined through the size and resolution of the preset area. The noise intensity of the noise in the CT image is calculated based on the CT value of the pixel and the number of pixels.

[0023] According to an embodiment of the present invention, determining the noise intensity of the noise in the CT image according to the CT value of the pixel and the number of the pixels includes: determining the noise intensity SD of the noise in the CT image according to formula (1), (1) in, is the CT value of the i-th pixel in the preset area, n is the number of pixels in the preset area, i≤n, and both i and n are positive integers.

[0024] According to one embodiment of the present invention, the noise in the CT image is usually manifested as the variation of the CT value in the image with respect to its average value. is the standard deviation of n pixels in a preset area in the CT image, indicating the fluctuation degree of the CT value of the pixels in the preset area, that is, the intensity of the noise. The larger the standard deviation, the greater the intensity of the noise.

[0025] In this way, the noise intensity of the noise in the CT image can be determined according to the CT value of the pixel and the number of pixels, which provides a data basis for the subsequent determination of the training data set and improves the calculation efficiency.

[0026] According to an embodiment of the present invention, in step S104, a training data set is determined according to the noise type and the noise intensity, wherein the training noise type of the training CT image in the training data set is the same as the noise type.

[0027] For example, according to the noise type and noise intensity of the CT image, noise with the same noise type and small difference in noise intensity is simulated and introduced in the training dataset.

[0028] According to an embodiment of the present invention, in step S105, the noise processing model is trained according to the training data set to obtain a trained noise processing model.

[0029] According to one embodiment of the present invention, step S105 includes: Determine the amount of data for each training data set; Determining a pixel value of each pixel in the training CT image; Determine the spatial resolution and contrast resolution of the training CT images; Acquire a high-quality CT image, and determine reference pixel values ​​of corresponding pixels in the high-quality CT image and the training CT image and a reference noise intensity of the high-quality CT image; Determining a peak signal-to-noise ratio according to the pixel value and the reference pixel value; Processing the training CT image through the noise processing model to obtain a processed training CT image; determining predicted noise intensity of processed training CT images; Determining a training loss function of a noise processing model according to the spatial resolution, the contrast resolution, the peak signal-to-noise ratio, the data volume, the predicted noise intensity, and the reference noise intensity; The noise processing model is trained according to the training loss function to obtain a trained noise processing model.

[0030] For example, each training data set contains multiple CT images for training, and the size of each training data set, that is, the data volume of the training data set, is determined; the pixel value of each pixel in the training CT image is obtained by a CT scanner; the spatial resolution and contrast resolution of the training CT image are detected by professional testing methods and tools (for example, a sine wave test card); a high-quality CT image is selected as a reference image for the training CT image, and the high-quality CT image can be an undamaged image, or an image that is considered to be closest to the actual situation after some preprocessing, and the high-quality CT image has the same resolution as the training CT image, that is, it has the same number of pixels and arrangement, and obtains a high-quality CT image. The invention relates to a method for detecting the noise intensity of a high-quality CT image and a reference pixel value of a corresponding pixel in a training CT image; calculating the peak signal-to-noise ratio according to the pixel value of the training CT image and the reference pixel value of the high-quality CT image; the noise processing model is a deep learning neural network model, which can process the CT image and output a denoised CT image; detecting the noise intensity of the processed training CT image, that is, predicting the noise intensity; determining the training loss function of the noise processing model according to the spatial resolution, contrast resolution, peak signal-to-noise ratio, data volume, predicted noise intensity and reference noise intensity; and training the noise processing model according to the trained loss function to obtain a trained noise processing model.

[0031] According to an embodiment of the present invention, determining the peak signal-to-noise ratio according to the pixel value and the reference pixel value includes: determining the peak signal-to-noise ratio of the kth training CT image in the jth training data set according to formula (2): , (2) Among them, max is the maximum value function, is the pixel value of the rth pixel of the kth training CT image in the jth training data set, is the pixel value of the rth corresponding pixel of the high-quality CT image corresponding to the kth training CT image in the jth training data set, R is the number of pixels in the training CT image, r≤R, and r and R are both positive integers.

[0032] According to one embodiment of the present invention, To obtain the maximum pixel value of the kth training CT image in the jth training data set, is the mean square error, which represents the average value of the square of the error between the signal and the noise. It represents the ratio of the maximum possible power to the destructive noise power, that is, the peak signal-to-noise ratio. The larger the peak signal-to-noise ratio, the smaller the noise in the CT image and the higher the image quality.

[0033] In this way, the peak signal-to-noise ratio can be determined based on the pixel value and the reference pixel value, thereby improving the convenience of calculation.

[0034] According to one embodiment of the present invention, the training loss function of the noise processing model is determined according to the spatial resolution, the contrast resolution, the peak signal-to-noise ratio, the data volume, the predicted noise intensity and the reference noise intensity, including: determining the training loss function of the noise processing model according to formula (3): , (3) in, is the peak signal-to-noise ratio of the kth training CT image in the jth training dataset, is the preset peak signal-to-noise ratio threshold, is the spatial resolution of the kth training CT image in the jth training dataset, is the preset spatial resolution threshold, is the contrast resolution of the kth training CT image in the jth training dataset, is the preset contrast resolution threshold, is the predicted noise intensity of the kth training CT image in the jth training dataset, is the reference noise intensity of the high-quality image of the k-th training CT image in the j-th training dataset, is the data size of the jth training data set, m is the number of training data sets, j≤m, K is the number of training CT images in the training data set, k≤K, j, m, k and K are all positive integers.

[0035] According to one embodiment of the present invention, is the ratio of the spatial resolution of the kth training CT image in the jth training data set to the preset spatial resolution threshold. The larger the ratio is, the higher the spatial resolution of the kth training CT image in the jth training data set is. is the ratio of the contrast resolution of the kth training CT image in the jth training data set to the preset contrast resolution threshold. The larger the ratio is, the higher the contrast resolution of the kth training CT image in the jth training data set is. is the ratio of the peak signal-to-noise ratio of the kth training CT image in the jth training data set to the preset peak signal-to-noise ratio threshold. The larger the ratio, the larger the peak signal-to-noise ratio of the kth training CT image in the jth training data set. It means that the spatial resolution and contrast resolution are positively correlated with the noise intensity, and the peak signal-to-noise ratio is negatively correlated with the noise intensity. For example, the higher the spatial resolution of the CT image, the less the number of received photons, and the greater the noise intensity. The higher the contrast resolution of the CT image, the higher the exposure conditions, the more X-ray photons are generated, resulting in an increase in noise intensity. The smaller the peak signal-to-noise ratio, the worse the image quality and the higher the noise intensity. It is the ratio of the data volume of the jth training data set to the average data volume of the m training data sets. The larger the ratio is, the larger the data volume of the jth training data set is, the greater the impact on the training process is, and the greater the impact of its change on the error of the predicted noise intensity is. is the relative error between the predicted noise intensity of the kth training CT image in the jth training dataset and the reference noise intensity of the high-quality image of the kth training CT image in the jth training dataset.

[0036] According to one embodiment of the present invention, using and The relative error of the predicted noise intensity of each training CT image is weighted averaged to obtain a training loss function. During the training process, the training loss function is reduced, so that the error between the predicted noise intensity and the reference noise intensity is reduced, and the noise processing model improves the accuracy of noise processing, thereby improving the accuracy of the noise processing model.

[0037] In this way, the influence of the above data on the error of predicting noise intensity can be determined according to the influence of spatial resolution, contrast resolution and peak signal-to-noise ratio on noise intensity, so as to set the training loss function based on this influence, the relative error of predicting noise intensity and the data amount of the training data set, so that the training loss function of the noise processing model can be reduced during the training process, and the trained noise processing model can be more suitable for noise reduction processing of CT images of the current type and intensity, thereby improving the accuracy of the noise processing model in a more targeted manner.

[0038] According to one embodiment of the present invention, training the noise processing model according to the training loss function to obtain the trained noise processing model includes: Determining a total training amount based on the data amount; Determining a first ratio according to the total training amount and a preset training amount threshold; Determining a training learning rate according to a preset learning rate threshold and the first ratio; The noise processing model is trained according to the training loss function and the training learning rate to obtain a trained noise processing model.

[0039] For example, large-scale data sets can provide rich samples and features, which help the model learn the wide distribution and potential patterns of the data. For large-scale data sets, a larger learning rate can usually be selected because a larger data set can provide stable gradient estimates, allowing the model to perform larger step parameter updates at each iteration. Small-scale data sets have limited sample sizes, which may lead to model overfitting or underfitting. For small-scale data sets, a smaller learning rate is usually used to avoid model instability or overfitting caused by excessive weight updates. Training data is summed up according to the data volume of each training data set to determine the total training volume; a first ratio is determined according to the ratio of the total training volume to a preset training volume threshold, indicating the scale of the data set. The larger the first ratio, the larger the scale of the data set, and a relatively larger learning rate should be selected; the training learning rate is determined according to the product of the first ratio and the preset learning rate threshold; the noise processing model is trained according to the training loss function and the training learning rate to obtain a trained noise processing model.

[0040] According to an embodiment of the present invention, in step S106, the CT image is processed according to the trained noise processing model to obtain a denoised CT image.

[0041] For example, by using the trained noise processing model to process CT images and restore clean images from them, the noise level in the CT images can be effectively reduced through noise reduction processing, thus improving the clarity of the image, helping to identify the details of the damaged area, and enhancing the ability to identify damage.

[0042] According to an embodiment of the present invention, in step S107, the lesion area at the lesion position in the noise-reduced CT image is determined, and the second CT value of the pixel at the lesion position is determined.

[0043] For example, the lesion area at the lesion position in the noise-reduced CT image and the CT value of the pixel at the lesion position, ie, the second CT value, are determined by a CT scanner.

[0044] According to an embodiment of the present invention, in step S108, an injury severity coefficient is determined according to the injury area and the second CT value.

[0045] According to one embodiment of the present invention, step S108 includes: Determine the damage centroid at the damage location; determining the damage shape of the edge at the damage location; Determine a first similar shape and a second similar shape according to the damage centroid and the damage shape, wherein the first similar shape and the second similar shape are centered on the damage centroid, and the first similar shape and the second similar shape are the same as the damage shape; Determine a first pixel sampling point at an edge of the first similar shape, and determine a second CT value of the first pixel sampling point; Determine a second pixel sampling point at an edge of the second similar shape, and determine a second CT value of the second pixel sampling point; An injury severity coefficient is determined according to the second CT value of the first pixel sampling point, the second CT value of the second pixel sampling point, and the injury area.

[0046] For example, the centroid (i.e., the centroid) and shape (i.e., the shape of the injury) of the injury position in the CT image are determined; based on the centroid and shape of the injury, a first similar shape and a second similar shape are determined; for example, if the shape of the injury at the edge of the injury position is a circle, then the first similar shape and the second similar shape are concentric circles with the centroid of the injury position as the center, the radius of the first similar shape is larger, and the radius of the second similar shape is smaller, and both the first similar shape and the second similar shape are located within the injury position, for example, if the injury position is a circle with a radius of r, then the first similar shape is a circle with a radius greater than 0.5r and less than r, and the second similar shape is a circle with a radius less than 0.5r; uniform sampling is performed at the edge of the first similar shape to determine a first pixel sampling point, and the CT value of the first pixel sampling point is detected; uniform sampling is performed at the edge of the second similar shape to determine a second pixel sampling point, and the CT value of the second pixel sampling point is detected; based on the second CT value of the first pixel sampling point, the second CT value of the second pixel sampling point, and the injury area, the injury degree of the injury part in the CT image is evaluated to determine the injury severity coefficient.

[0047] According to an embodiment of the present invention, determining the injury severity coefficient according to the second CT value of the first pixel sampling point, the second CT value of the second pixel sampling point and the injury area includes: determining the injury severity coefficient according to formula (4): , (4) Among them, max is the maximum value function, , and is the preset weight, is the damaged area, is the second CT value of the fth first pixel sampling point, is the second CT value of the f-th second pixel sampling point, is the preset CT value, F is the number of the first pixel sampling points, f≤F, and both f and F are positive integers.

[0048] According to one embodiment of the present invention, Indicates the relative difference between the second CT value of the f-th first pixel sampling point and the preset CT value. The larger the ratio, the greater the change of the second CT value of the f-th first pixel sampling point relative to the preset CT value, and the greater the degree of damage at the f-th first pixel sampling point. When a part is damaged (such as fracture, bleeding, inflammation, etc.), the density of the part may change, resulting in a change in the CT value. When the number of first pixel sampling points with a higher degree of damage in the first similar shape is large, the preset weight corresponding to the first pixel sampling point is The larger the value of It means taking the maximum value of the sum of the products of the damage degree and the corresponding weight of the F first pixel sampling points in the first similar shape. For example, when the first first pixel sampling point in the first similar shape is The value of is 1, the value at the second first pixel sampling point is 0.8, and the value at the second first pixel sampling point is 0.9. The preset weights are , and ,and ,but The value of , which means that when there are more first pixel sampling points with higher damage degree in the first similar shape, the impact of the damage degree on the edge of the first similar shape becomes more serious as the number of first pixel sampling points with higher damage degree increases, wherein the preset weight The change of is a combination of exponential function and linear function. The part of the change in the form of a linear function indicates that the damage degree at the single first pixel sampling point increases, causing the overall damage degree at the edge of the first similar shape to increase at a uniform rate. The exponential function part of the change in indicates that as the number of the first pixel sampling points with higher damage increases, the impact on the overall damage degree at the edge of the first similar shape increases rapidly, and it does not increase at a uniform rate. The above method of allocating weight values ​​to find the maximum value can be used to solve the most serious overall damage condition at the edge of the first similar shape, that is, the overall damage condition of the peripheral area in the damage position, which is convenient for evaluating the severity of the patient's condition. Similarly, It can be used to solve the situation where the overall damage degree at the edge of the second similar shape is the most serious, that is, the overall damage condition of the central area in the damage position, To determine the overall damage degree of the damage location according to the overall damage degree of the peripheral area in the damage location and the overall damage degree of the central area in the damage location, wherein: Less than , indicating that the overall damage degree of the central area of ​​the damage location has a greater impact on the overall damage degree of the damage location. To determine the comprehensive damage severity of the damage location based on the overall damage degree and damage area of ​​the damage location.

[0049] In this way, the damage severity coefficient can be determined based on the second CT value of the first pixel sampling point, the second CT value of the second pixel sampling point and the damage area. During the calculation process, the most serious overall damage level of the peripheral area in the damage position and the central area in the damage position can be determined based on the damage level at the pixel sampling points in the similar shape. Furthermore, the overall damage level of the damage position is determined based on the overall damage level of the peripheral area in the damage position and the central area in the damage position, and the comprehensive damage severity level of the damage position is determined in combination with the damage area, thereby improving the comprehensiveness and accuracy of the damage severity coefficient.

[0050] According to an embodiment of the present invention, in step S109, an injury report is generated according to the injury severity coefficient.

[0051] For example, based on the damage severity coefficient, the damage severity at the damage location is determined and a damage report is generated.

[0052] According to the medical injury CT image processing method of the embodiment of the present invention, the noise type and noise intensity in the medical injury CT image can be accurately analyzed, and the training data set is determined according to the noise type and noise intensity to train the noise processing model, and the injury area and CT value of the injury part are determined according to the medical injury CT image after noise reduction. Further, the degree of injury at the injury position is evaluated according to the injury area and CT value, which is conducive to improving the accuracy and effectiveness of the detailed analysis of the injury position in the CT image. When determining the noise intensity, the noise intensity of the noise in the CT image can be determined according to the CT value of the pixel and the number of pixels, which provides a data basis for the subsequent determination of the training data set and improves the calculation efficiency. When determining the peak signal-to-noise ratio, the peak signal-to-noise ratio can be determined according to the pixel value and the reference pixel value, which improves the convenience of calculation. When determining the training loss function, the influence of the spatial resolution, contrast resolution and peak signal-to-noise ratio on the noise intensity can be used to determine the influence of the above data on the error of predicting the noise intensity, so as to set the training loss function based on the influence, the relative error of predicting the noise intensity and the data volume of the training data set, so that the noise processing model can reduce the training loss function during the training process, and make the trained noise processing model more suitable for the noise reduction processing of the current type and intensity of CT images, and improve the accuracy of the noise processing model in a more targeted manner. When determining the injury severity coefficient, the injury severity coefficient is determined according to the second CT value of the first pixel sampling point, the second CT value of the second pixel sampling point and the injury area. In the calculation process, the most serious overall injury degree of the peripheral area in the injury position and the central area in the injury position can be determined according to the injury degree at the pixel sampling point in the similar shape. Further, the overall injury degree of the injury position is determined according to the overall injury degree of the peripheral area in the injury position and the central area in the injury position, and the comprehensive injury severity of the injury position is determined in combination with the injury area, thereby improving the comprehensiveness and accuracy of the injury severity coefficient.

[0053] Figure 2 A block diagram of a medical injury CT image processing system according to an embodiment of the present invention is exemplarily shown, wherein the system comprises: A preprocessing module, used for performing data preprocessing on CT images; A noise type module, used to determine the noise type of noise in the CT image; A noise intensity module for determining the noise intensity of noise in the CT image; A training data module, used to determine a training data set according to the noise type and the noise intensity, wherein the training noise type of the training CT image in the training data set is the same as the noise type; A training model module, used to train the noise processing model according to the training data set to obtain a trained noise processing model; A training model module, used to process the CT image according to the trained noise processing model to obtain a denoised CT image; An image analysis module, used to determine the damage area at the damage position in the CT image after noise reduction, and determine the second CT value of the pixel at the damage position; An injury severity module, used to determine an injury severity coefficient according to the injury area and the second CT value; The damage report module is used to generate a damage report according to the damage severity coefficient.

[0054] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0055] It should be understood by those skilled in the art that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

Claims

1. A method for processing medical injury CT images, characterized in that: include: Perform data preprocessing on CT images; Determine the noise type of noise in CT images; Determine the noise intensity of noise in CT images; Determining a training data set according to the noise type and the noise intensity, wherein the training noise type of the training CT image in the training data set is the same as the noise type; Training a noise processing model according to the training data set to obtain a trained noise processing model; Processing the CT image according to the trained noise processing model to obtain a denoised CT image; Determine the lesion area at the lesion location in the noise-reduced CT image, and determine a second CT value of the pixel at the lesion location; determining an injury severity coefficient according to the injury area and the second CT value; An injury report is generated based on the injury severity coefficient.

2. The medical injury CT image processing method according to claim 1, characterized in that: Determine the noise intensity of the noise in the CT image, including: In the CT image, a preset area is determined; Obtaining the CT value and number of pixels of each pixel in a preset area; The noise intensity of the noise in the CT image is determined according to the CT value of the pixel and the number of the pixels.

3. The medical injury CT image processing method according to claim 2, characterized in that: Determining the noise intensity of noise in the CT image according to the CT value of the pixel and the number of the pixels includes: According to the formula ； Determine the noise intensity SD of the noise in the CT image, where, is the CT value of the i-th pixel in the preset area, n is the number of pixels in the preset area, i≤n, and both i and n are positive integers.

4. The medical injury CT image processing method according to claim 1, characterized in that: According to the training data set, the noise processing model is trained to obtain a trained noise processing model, including: Determine the amount of data for each training data set; Determining a pixel value of each pixel in the training CT image; Determine the spatial resolution and contrast resolution of the training CT images; Acquire a high-quality CT image, and determine reference pixel values ​​of corresponding pixels in the high-quality CT image and the training CT image and a reference noise intensity of the high-quality CT image; Determining a peak signal-to-noise ratio according to the pixel value and the reference pixel value; Processing the training CT image through the noise processing model to obtain a processed training CT image; determining predicted noise intensity of processed training CT images; Determining a training loss function of a noise processing model according to the spatial resolution, the contrast resolution, the peak signal-to-noise ratio, the data volume, the predicted noise intensity, and the reference noise intensity; The noise processing model is trained according to the training loss function to obtain a trained noise processing model.

5. The medical injury CT image processing method according to claim 4, characterized in that: Determining a peak signal-to-noise ratio according to the pixel value and the reference pixel value includes: According to the formula ； Determine the peak signal-to-noise ratio of the kth training CT image in the jth training dataset , where max is the maximum value function, is the pixel value of the rth pixel of the kth training CT image in the jth training data set, is the pixel value of the rth corresponding pixel of the high-quality CT image corresponding to the kth training CT image in the jth training data set, R is the number of pixels in the training CT image, r≤R, and r and R are both positive integers.

6. The medical injury CT image processing method according to claim 4, characterized in that: Determining a training loss function of a noise processing model according to the spatial resolution, the contrast resolution, the peak signal-to-noise ratio, the data volume, the predicted noise intensity, and the reference noise intensity includes: According to the formula ； Determine the training loss function for the noise processing model ,in, is the peak signal-to-noise ratio of the kth training CT image in the jth training dataset, is the preset peak signal-to-noise ratio threshold, is the spatial resolution of the kth training CT image in the jth training dataset, is the preset spatial resolution threshold, is the contrast resolution of the kth training CT image in the jth training dataset, is the preset contrast resolution threshold, is the predicted noise intensity of the kth training CT image in the jth training dataset, is the reference noise intensity of the high-quality image of the k-th training CT image in the j-th training dataset, is the data size of the jth training data set, m is the number of training data sets, j≤m, K is the number of training CT images in the training data set, k≤K, j, m, k and K are all positive integers.

7. The medical injury CT image processing method according to claim 4, characterized in that: The noise processing model is trained according to the training loss function to obtain a trained noise processing model, including: Determining a total training amount based on the data amount; Determining a first ratio according to the total training amount and a preset training amount threshold; Determining a training learning rate according to a preset learning rate threshold and the first ratio; The noise processing model is trained according to the training loss function and the training learning rate to obtain a trained noise processing model.

8. The medical injury CT image processing method according to claim 1, characterized in that: Determining an injury severity coefficient according to the injury area and the second CT value includes: Determine a damage centroid at the damage location, wherein the damage centroid is the centroid of the area where the damage location is located; determining the damage shape of the edge at the damage location; Determine a first similar shape and a second similar shape according to the damage centroid and the damage shape, wherein the first similar shape and the second similar shape are centered on the damage centroid, and the first similar shape and the second similar shape are the same as the damage shape; Determine a first pixel sampling point at an edge of the first similar shape, and determine a second CT value of the first pixel sampling point; Determine a second pixel sampling point at an edge of the second similar shape, and determine a second CT value of the second pixel sampling point; An injury severity coefficient is determined according to the second CT value of the first pixel sampling point, the second CT value of the second pixel sampling point, and the injury area.

9. The medical injury CT image processing method according to claim 8, characterized in that: Determining the injury severity coefficient according to the second CT value of the first pixel sampling point, the second CT value of the second pixel sampling point, and the injury area includes: According to the formula ； Determining the Injury Severity Factor , where max is the maximum value function, , and is the preset weight, is the damaged area, is the second CT value of the fth first pixel sampling point, is the second CT value of the f-th second pixel sampling point, is the preset CT value, F is the number of the first pixel sampling points, f≤F, and both f and F are positive integers.

10. A medical injury CT image processing system, characterized in that: include: A preprocessing module, used for performing data preprocessing on CT images; A noise type module, used to determine the noise type of noise in the CT image; A noise intensity module for determining the noise intensity of noise in the CT image; A training data module, used to determine a training data set according to the noise type and the noise intensity, wherein the training noise type of the training CT image in the training data set is the same as the noise type; A training model module, used to train the noise processing model according to the training data set to obtain a trained noise processing model; A training model module, used to process the CT image according to the trained noise processing model to obtain a denoised CT image; An image analysis module, used to determine the damage area at the damage position in the CT image after noise reduction, and determine the second CT value of the pixel at the damage position; An injury severity module, used to determine an injury severity coefficient according to the injury area and the second CT value; The damage report module is used to generate a damage report according to the damage severity coefficient.

Images