Chest CT image processing method

Abstract

The invention relates to a chest CT (Computed Tomography) image processing method, which relates to the field of image processing, and comprises the following steps of: dividing a scanning time zone into Q expiration time zones and Q inspiration time zones on the basis of a respiration monitoring record, dividing, clustering and marking a chest CT image to be processed, and obtaining a 2Q local CT image set; constructing Q call artifact elimination plug-ins and Q suction artifact elimination plug-ins based on the convolutional neural network; and carrying out artifact mapping elimination on the 2Q local CT image set to obtain a 2Q standard local CT image set, and fusing to obtain a reconstructed chest CT image. According to the method, the technical problem that the breathing artifacts in the image cannot be accurately eliminated due to the fact that the difference of different breathing stages is difficult to capture on the fine-grained level in a traditional image processing method can be solved; the image is divided into a plurality of breathing time zones, and the artifacts are eliminated by using the deep learning model, so that the fineness and the accuracy of eliminating the breathing artifacts can be remarkably improved, and the quality and the precision of the chest CT image are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a chest CT image processing method. Background Art

[0002] In traditional chest CT scans, the patient's breathing cycle causes dynamic changes in the lungs during inhalation and exhalation. Such dynamic changes usually cause artifacts in the image. These artifacts often appear as blurred or striped shadows in the image, affecting the clarity and detail of the CT image. This is especially true in the detection of small lung nodules or early lesions. The presence of artifacts may lead to misdiagnosis or missed diagnosis.

[0003] Because patients go through different respiratory stages during the scanning process, traditional methods are often unable to specifically address each stage, resulting in the inability to accurately eliminate respiratory artifacts in the image. At the same time, respiratory artifacts will affect the clarity of the image, leading to inaccurate diagnosis of lung diseases and increasing the risk of misdiagnosis. Summary of the invention

[0004] The present invention provides a chest CT image processing method to solve the technical problem that traditional chest CT image processing methods are difficult to capture the differences between different respiratory stages at a fine-grained level, resulting in the inability to accurately eliminate respiratory artifact interference in images.

[0005] The technical solution of the present invention to solve the above technical problems is as follows: The present invention provides a chest CT image processing method, comprising: obtaining a chest CT image to be processed, and a layer-by-layer moving scanning record and a respiratory monitoring record during CT scanning; based on the respiratory monitoring record, dividing the scanning time zone into Q exhalation time zones and Q inhalation time zones, dividing, clustering and marking the chest CT image to be processed according to the Q exhalation time zones and the Q inhalation time zones, and obtaining a 2Q local CT image set; constructing a respiratory artifact elimination model based on a convolutional neural network, wherein the respiratory artifact elimination model includes Q exhalation artifact elimination plug-ins and Q inhalation artifact elimination plug-ins; using the Q exhalation artifact elimination plug-ins and the Q inhalation artifact elimination plug-ins, respectively performing respiratory artifact mapping elimination on the 2Q local CT image set to obtain a 2Q standard local CT image set, and after fusion, obtaining a reconstructed chest CT image as an image processing result.

[0006] Preferably, the chest CT image processing method further includes: when the user performs a chest CT scan, synchronously collecting the user's breathing characteristics according to a predetermined timestamp through a chest strap sensor, and obtaining a breathing feature sequence as a breathing monitoring record; based on the breathing monitoring record, dividing the scanning time zone according to Q exhalation stages and Q inhalation stages, and obtaining Q exhalation time zones and Q inhalation time zones.

[0007] Preferably, the chest CT image processing method further includes: subdividing the user's breathing cycle into Q exhalation phases and Q inhalation phases according to the chest fluctuation ratio, wherein Q is an integer greater than 2, the chest expansion ratio of each exhalation phase is different, and the chest contraction ratio of each inhalation phase is different.

[0008] Preferably, the chest CT image processing method also includes: aligning the respiratory monitoring records and the layer-by-layer motion scanning records in the time domain, dividing the chest CT images to be processed according to the Q exhalation time zones and the Q inhalation time zones, and clustering the local CT images of the same exhalation time zone or the same inhalation time zone; marking the clustering results according to the Q exhalation stages and the Q inhalation stages to obtain Q exhalation stage local CT image sets and Q inhalation stage local CT image sets, and combining them to obtain 2Q local CT image sets.

[0009] Preferably, the chest CT image processing method also includes: randomly selecting a first exhalation stage from Q exhalation stages, and obtaining a first chest expansion ratio of the first exhalation stage; using the first chest expansion ratio as a feature constraint, retrieving a first training set to perform supervised training on a convolutional neural network, and obtaining a first exhalation artifact elimination plug-in; based on the first exhalation artifact elimination plug-in, respectively analyzing the Q exhalation stages and the Q inhalation stages, obtaining Q exhalation artifact elimination plug-ins and Q inhalation artifact elimination plug-ins, and combining them to construct a respiratory artifact elimination model.

[0010] Preferably, the chest CT image processing method also includes: taking the first chest expansion ratio as a feature constraint, retrieving chest CT scan logs in a historical time zone, collecting sample first local CT images of multiple users, and obtaining sample first standard local CT images of multiple users, to obtain a sample first local CT image set and a sample first standard local CT image set, wherein the standard local CT image is a chest CT image without respiratory interference; using the sample first local CT image set and the sample first standard local CT image set as sample data sets, and dividing them into K equal parts to obtain K training sets, wherein K is an integer greater than 10; using the K training sets to perform supervised training on the convolutional neural network respectively until the model converges, to obtain K first call artifact elimination branches, and integrating to construct the first call artifact elimination plug-in.

[0011] Preferably, the chest CT image processing method also includes: randomly selecting a first local CT image set in the first exhalation stage or the first inhalation stage from the 2Q local CT image set, and matching to obtain a first artifact elimination plug-in, wherein the first artifact elimination plug-in is a first exhalation artifact elimination plug-in or a first inhalation artifact elimination plug-in; setting the number P of adaptation branch selections based on the user's body weight index, wherein P is less than or equal to K; randomly selecting P artifact elimination branches from the K artifact elimination branches of the first artifact elimination plug-in, performing artifact elimination on the first local CT image set, and outputting a first standard local CT image set after mean calculation, and adding it to the 2Q standard local CT image set.

[0012] Preferably, the chest CT image processing method also includes: obtaining the user's body mass index, and calculating the deviation ratio between the user's body mass index and a preset standard body mass index to obtain an individual deviation coefficient; multiplying 1 and the sum of the individual deviation coefficient by a predetermined selection number and rounding the result to obtain P, where the predetermined selection number is 5.

[0013] The beneficial effects of the present invention are as follows: by obtaining a chest CT image to be processed, and a layer-by-layer moving scanning record and a respiratory monitoring record during CT scanning; then based on the respiratory monitoring record, the scanning time zone is divided into Q breathing time zones and Q breathing time zones, and the chest CT image to be processed is divided, clustered and identified according to the Q breathing time zones and the Q breathing time zones, and a 2Q local CT image set is obtained; then a respiratory artifact elimination model is constructed based on a convolutional neural network, wherein the respiratory artifact elimination model includes Q breathing artifact elimination plug-ins and Q breathing artifact elimination plug-ins; finally, the Q breathing artifact elimination plug-ins and the Q breathing artifact elimination plug-ins are used to respectively eliminate respiratory artifact mapping of the 2Q local CT image set, and a 2Q standard local CT image set is obtained after fusion, and a reconstructed chest CT image is obtained as an image processing result. That is to say, by dividing the image into multiple breathing time zones and using a deep learning model to eliminate artifacts, the fineness and accuracy of respiratory artifact elimination can be significantly improved, thereby effectively improving the quality and accuracy of chest CT images. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic diagram of a process flow of a chest CT image processing method provided by the present invention; Figure 2 A schematic diagram of a process for obtaining a 2Q local CT image set in a chest CT image processing method provided by the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0016] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0017] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or explanation". Any embodiment described as "for example" in the present invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present invention.

[0018] Examples, such as Figure 1 As shown, the embodiment of the present invention provides a chest CT image processing method, which specifically includes the following steps: S10: Acquire the chest CT image to be processed, as well as the layer-by-layer motion scanning record and the respiratory monitoring record during the CT scanning.

[0019] Specifically, the chest CT image to be processed is obtained. The chest CT image to be processed refers to the medical image obtained by a chest CT scanner, which includes cross-sectional images of various areas of the patient's chest and has not undergone image optimization processing; on the other hand, the layer-by-layer motion scanning record and respiratory monitoring record during the CT scan are obtained. The layer-by-layer motion scanning record refers to the CT scanner recording the specific scanning position, layer thickness, scanning time and other information of each image slice during the scanning process. These records are helpful for subsequent image reconstruction and analysis, especially when precise image processing is required. The layer-by-layer scanning record can provide the acquisition time point and position of each layer of the image; the respiratory monitoring record is a real-time collection of the patient's respiratory data at various stages of the CT scanning process through a special respiratory monitoring device. These data include the amplitude of the user's chest movement during the scanning process (that is, the degree of chest expansion and contraction). These records can help analyze the patient's breathing pattern and are used to divide CT images into different respiratory stages for targeted processing.

[0020] During the image processing stage, layer-by-layer motion scanning records and respiratory monitoring records are used together with the original CT images. These data can help doctors or image processing systems accurately understand the patient's respiratory status during the scan, and provide key support for subsequent artifact elimination, image reconstruction and other processing. For example, respiratory monitoring records can identify the time correspondence between each stage of the CT image (inhalation, exhalation) and the specific image data, thereby accurately eliminating artifacts caused by breathing.

[0021] S20: Based on the respiratory monitoring record, the scanning time zone is divided into Q exhalation time zones and Q inhalation time zones, and the chest CT image to be processed is divided, clustered and marked according to the Q exhalation time zones and Q inhalation time zones to obtain a 2Q local CT image set.

[0022] Further, step S20 of the present invention further includes: S21: When the user undergoes a chest CT scan, the chest strap sensor synchronously collects the user's breathing characteristics according to a predetermined time stamp to obtain a breathing characteristic sequence as a breathing monitoring record.

[0023] Specifically, a chest strap sensor is a device used to monitor the user's chest movement. It is usually equipped with a pressure sensor or an accelerometer, which can sense and record the expansion and contraction of the user's chest, thereby reflecting the breathing process. The chest strap is usually worn on the chest cavity area of ​​the chest, and can obtain the user's inhalation and exhalation status in real time. Among them, the chest strap sensor generates a pressure signal corresponding to the respiratory cycle by detecting the expansion and contraction of the user's chest (i.e., the change in chest volume during inhalation and exhalation); through the accelerometer, the chest strap can record the amplitude and frequency of chest movement in real time, so as to calculate the user's respiratory frequency, duration of inhalation and exhalation, and other characteristics. In the process of chest CT scanning, accurately capturing the user's respiratory state is crucial to improving image quality and subsequent image processing (such as artifact removal). By combining the synchronous acquisition mechanism of the chest strap sensor and the CT scanner, the user's respiratory characteristics can be obtained in real time, thereby providing the necessary respiratory monitoring data for subsequent fine image processing.

[0024] Next, when the user undergoes a chest CT scan, the chest strap sensor synchronously collects the user's breathing characteristics according to a predetermined timestamp. During the CT scan, a timestamp is usually set for each image slice (for example, recording the specific time of the scan). In order to effectively pair the breathing characteristic data with the chest CT image, the chest strap sensor must synchronize the collection with the CT scan process. By setting a predetermined timestamp, the chest strap sensor can record the user's breathing characteristics at each time point of CT image acquisition; then, based on the pressure signal and acceleration data recorded by the chest strap sensor, a breathing characteristic sequence is generated. This sequence includes the user's inhalation and exhalation phase information, chest movement amplitude, and the start and end time of the breathing cycle at each time point during the CT scan.

[0025] The chest strap sensor synchronously collects the user's breathing characteristics during the chest CT scan process. Combined with a predetermined timestamp, it can generate a detailed breathing feature sequence. These sequences serve as respiratory monitoring records and provide accurate respiratory status data for subsequent image processing. With these data, breathing artifacts in the image can be eliminated at a fine-grained level, ultimately improving the quality of CT images and diagnostic accuracy.

[0026] S22: Based on the respiratory monitoring record, the scanning time zone is divided according to Q exhalation phases and Q inhalation phases to obtain Q exhalation time zones and Q inhalation time zones.

[0027] Further, step S22 of the present invention further includes: S221: According to the chest fluctuation ratio, the user's breathing cycle is subdivided into Q exhalation phases and Q inhalation phases, where Q is an integer greater than 2, the chest expansion ratio of each exhalation phase is different, and the chest contraction ratio of each inhalation phase is different.

[0028] Specifically, in the process of chest CT image processing, considering the impact of breathing on the chest cavity, in order to more accurately eliminate respiratory artifacts and improve image quality, the user's respiratory cycle can be finely divided according to the chest fluctuation ratio. The respiratory cycle refers to the complete cycle from the beginning of one inhalation to the end of the next inhalation, usually including two major stages of inhalation and exhalation. In actual operation, the respiratory cycle is not completely uniform, and its expansion and contraction process has different speeds and intensities in different stages; in order to accurately capture the details of different respiratory stages and ensure the accuracy of subsequent image processing, the respiratory cycle can be divided into multiple smaller stages according to the chest fluctuation ratio. The division of each stage is dynamically allocated according to the patient's chest expansion and contraction amplitude, so that the image of each stage can be processed more accurately.

[0029] The chest fluctuation ratio is obtained. The chest fluctuation ratio refers to the ratio of chest changes in different stages of chest expansion (inhalation) and contraction (exhalation) during the breathing process. These ratios are usually manifested as changes in chest expansion and contraction, which can be recorded in real time by a chest strap sensor or other respiratory monitoring equipment. The chest fluctuation ratio can be set according to demand, for example, 20%. In order to further improve the accuracy of image processing, the entire respiratory cycle is divided into Q exhalation stages and Q inhalation stages according to the chest fluctuation ratio, that is, the inhalation process is subdivided into Q sub-stages, and the chest expansion ratio of each sub-stage is different, and the exhalation process is subdivided into Q sub-stages, and the chest contraction ratio of each sub-stage is different, where Q is an integer greater than 2. For example, assuming that the chest fluctuation ratio is 20%, Q is 5. By subdividing the user's breathing cycle into Q exhalation phases and Q inhalation phases, and processing according to the chest expansion and contraction ratios in each phase, more refined image analysis and artifact elimination can be achieved. This method effectively improves the accuracy and clarity of chest CT images. For example, in certain phases, the patient's breathing may be more rapid or deep, and strong artifacts may appear in the image at this time. Clear division of phases helps to accurately eliminate these artifacts.

[0030] Then, based on the respiratory monitoring record, the scanning time zone is divided according to Q exhalation stages and Q inhalation stages, that is, according to the timestamp and respiratory stage in the respiratory monitoring record, the scanning time zone can be divided into Q exhalation time zones and Q inhalation time zones, and Q exhalation time zones and Q inhalation time zones are obtained, wherein the exhalation time zones represent different stages of inhalation (exhalation stages) during the breathing process, and each exhalation time zone corresponds to a specific exhalation time period. In each exhalation time zone, the patient's chest expansion ratio is different, so the respiratory artifacts affecting the image quality are also different; the inhalation time zone represents different stages of exhalation (inhalation stage), and each inhalation time zone corresponds to a specific inhalation time period. The degree of contraction of the patient's chest is different, which affects the artifact performance in the image.

[0031] Based on the respiratory monitoring records, the scanning time zone is divided into Q exhalation time zones and Q inhalation time zones. The corresponding image processing task can be assigned to each respiratory stage. Through the refined division of respiratory stages and synchronization with the CT image time, the accuracy and effect of respiratory artifact elimination can be improved, and ultimately the quality of chest CT images can be improved.

[0032] Further, if Figure 2 As shown, step S20 of the present invention further includes: S23: Align the respiratory monitoring records and the layer-by-layer motion scanning records in the time domain, divide the chest CT images to be processed according to the Q exhalation time zones and the Q inhalation time zones, and cluster the local CT images of the same exhalation time zone or the same inhalation time zone; S24: Perform stage identification on the clustering results according to the Q exhalation stages and the Q inhalation stages, obtain Q exhalation stage local CT image sets and Q inhalation stage local CT image sets, and combine them to obtain 2Q local CT image sets.

[0033] Specifically, the respiratory monitoring record and the layer-by-layer motion scanning record are aligned in the time domain. Time domain alignment refers to accurately matching the respiratory monitoring record with the layer-by-layer motion scanning record in time. Since the chest CT scan is scanned layer by layer, the acquisition of each layer of the image has a timestamp, and the respiratory monitoring record will also record the patient's inhalation and exhalation information at each stage. In order to ensure that the two records can accurately correspond on the same timeline, time domain alignment is required. Through time domain alignment, each time period of the respiratory monitoring record (such as the inhalation or exhalation stage) will correspond to the acquisition time period of each layer of the CT scan. The data after time domain alignment provides accurate time matching for subsequent steps, so that there is a clear correspondence between each respiratory stage and the scanned image.

[0034] Next, after alignment in the time domain, the scan records are divided into Q exhalation time zones and Q inhalation time zones. Each time zone corresponds to a different respiratory stage, so the CT images can be subdivided and processed; each exhalation time zone corresponds to an exhalation stage, and the changes of the CT images in these stages will vary due to different chest expansions; each inhalation time zone corresponds to an inhalation stage, and the changes of the CT images in these stages will vary due to different chest contractions. Then, the local CT images of the same exhalation time zone or the same inhalation time zone are clustered, that is, image slices with similar features are grouped together, thereby improving the efficiency and accuracy of subsequent processing. For example, in the same exhalation time zone, all images may show similar chest expansion and breathing patterns, so their features will be identified as similar in the clustering process, so that they can be processed together. Similarly, all images in the inhalation time zone can also be divided into multiple groups by clustering methods.

[0035] Then, the clustering results are stage-labeled according to the Q exhalation phases and Q inhalation phases, that is, each image group in the clustering results is labeled as the corresponding respiratory phase. This process is to ensure that each image set can be accurately identified and operated in subsequent processing. By labeling each clustering result, the image is finally divided into Q exhalation phase local CT image sets and Q inhalation phase local CT image sets. Each image set contains all local CT images in this phase, and these images have similar breathing patterns and chest expansion / contraction features. Among them, each exhalation phase image set contains images of different depths and intensities during exhalation, and each inhalation phase image set contains images of different depths and intensities during inhalation. Finally, the Q exhalation phase local CT image sets and the Q inhalation phase local CT image sets are combined to obtain 2Q local CT image sets. These local CT image sets provide accurate image data for subsequent respiratory artifact removal and image quality improvement; through this step, the CT image is refined and the image features of each stage are clearly labeled, thereby providing high-quality data input for the subsequent deep learning model (such as the artifact removal model).

[0036] S30: constructing a breathing artifact elimination model based on a convolutional neural network, wherein the breathing artifact elimination model includes Q exhalation artifact elimination plug-ins and Q inhalation artifact elimination plug-ins.

[0037] Further, step S30 of the present invention further includes: S31: randomly selecting a first breathing stage from the Q breathing stages, and obtaining a first chest expansion ratio of the first breathing stage.

[0038] Specifically, any one of the Q breathing stages is randomly selected as the first breathing stage, and a first chest expansion ratio of the first breathing stage is obtained, for example, a chest expansion of 40%.

[0039] S32: Using the first chest expansion ratio as a feature constraint, retrieve a first training set to perform supervised training on a convolutional neural network, and obtain a first breathing artifact elimination plug-in.

[0040] Further, step S32 of the present invention further includes: S321: Taking the first chest expansion ratio as a feature constraint, retrieve the chest CT scan log in the historical time zone, collect sample first local CT images of multiple users, and obtain sample first standard local CT images of multiple users, to obtain a sample first local CT image set and a sample first standard local CT image set, wherein the standard local CT image is a chest CT image without respiratory interference; S322: Use the sample first local CT image set and the sample first standard local CT image set as sample data sets, and divide them into K equal parts to obtain K training sets, wherein K is an integer greater than 10; S323: Use the K training sets to perform supervised training on the convolutional neural network respectively until the model converges, to obtain K first call artifact elimination branches, and integrate to construct the first call artifact elimination plug-in.

[0041] Specifically, first, the first chest expansion ratio is used as a feature constraint, which means that the system further limits the scope of subsequent retrieval operations based on the data obtained from the patient's respiratory state (such as the expansion ratio of the first exhalation phase). The first chest expansion ratio can be used to determine which historical data or user samples best match the current user's respiratory state; the first chest expansion ratio (such as 40%) is a description of the user's inhalation phase. This ratio represents the degree of expansion of the chest and can affect the degree of artifacts in the image. Therefore, through this ratio, historical chest CT data similar to the current user's respiratory state can be filtered out.

[0042] Next, based on the first chest expansion ratio, the chest CT scan logs in the historical time zone are retrieved. The historical time zone refers to the records of CT scans that have been performed before. These records contain CT image data of different users at different breathing stages. These data may come from scan records of different time periods, covering CT scan information of multiple users at similar breathing expansion ratios. The goal of the retrieval is to find the historical data that is most similar to the current expansion ratio. Then, from the retrieved historical data, sample first local CT images of multiple users are collected. These images are selected from images matching the first expansion ratio in the historical time zone, and usually belong to local CT slices during the inhalation process; further, standard local CT images corresponding to these sample first local CT images are obtained from historical records. Standard local CT images refer to images without respiratory interference during the scanning process, representing ideal CT images without artifact interference; and a sample first local CT image set and a sample first standard local CT image set are obtained.

[0043] Then, the sample first local CT image set and the sample first standard local CT image set obtained in the previous step are merged to form a complete sample data set, which contains CT images of multiple users in different breathing states and their corresponding standardized images. This data set provides rich samples for model training and can support the learning of deep learning models. The sample data set is further divided into K parts, each of which contains a part of the sample data set, where K is an integer greater than 10 (usually 10, 20, etc.).

[0044] Convolutional neural network is a kind of deep learning model, which is widely used in image processing, video analysis, natural language processing and other fields, especially in computer vision tasks. CNN can automatically learn the spatial hierarchical features of input data, so it has significant advantages in image classification, object recognition, image segmentation and other tasks; then, K first call artifact elimination branches are constructed based on the convolutional neural network, and the first call artifact elimination branch includes input layer, convolution layer, pooling layer, fully connected layer and output layer; the input layer is used to receive input data, that is, the local CT image with breathing artifacts (the first local CT image of the sample); the convolution layer is responsible for extracting local features from the input image, such as edges, textures and more complex shape features in the image; the pooling layer usually follows the convolution layer, and its function is to reduce the size of the feature map, thereby reducing the amount of calculation and preventing overfitting; the fully connected layer maps the image features to the final output; the function of the output layer is to output a standard image corresponding to the input image, that is, the local CT image after artifacts are eliminated.

[0045] Then, the first local CT image of the sample is used as input, the first standard local CT image of the sample is used as supervision, and the K training sets are used to perform supervised training on the K first-call artifact elimination branches. The training goal is to restore the standard image from the local CT image with breathing artifacts through convolutional neural network learning. In supervised training, the network calculates the error between the predicted output (the image after artifact elimination) and the standard image, and adjusts the network according to the error; the commonly used loss function is the mean square error (MSE), which optimizes the network weights by minimizing the pixel difference between the input image and the target standard image; first, supervised training is performed on each CNN model using the training set; then, the error (loss function) between the output image and the standard image is calculated, and the network parameters are updated through the back propagation algorithm; the training process continues until the model achieves the best performance on the validation set, converges and stops training, and obtains K first-call artifact elimination branches; finally, the K first-call artifact elimination branches are integrated and combined to construct the first-call artifact elimination plug-in.

[0046] S33: Based on the first exhalation artifact elimination plug-in, the Q exhalation stages and the Q inhalation stages are analyzed respectively to obtain Q exhalation artifact elimination plug-ins and Q inhalation artifact elimination plug-ins, which are combined to construct a respiratory artifact elimination model.

[0047] Specifically, based on the same method of constructing the first exhalation artifact elimination plug-in, the images of Q exhalation stages and Q inhalation stages are analyzed respectively, that is, by training different exhalation stages, Q exhalation artifact elimination plug-ins are obtained, and each plug-in specializes in processing a specific exhalation stage; similarly, for different inhalation stages, Q inhalation artifact elimination plug-ins are obtained, and each plug-in eliminates the artifacts of an inhalation stage; finally, the obtained Q exhalation artifact elimination plug-ins and Q inhalation artifact elimination plug-ins are combined to form a complete respiratory artifact elimination model.

[0048] Through the analysis and combination of Q exhalation artifact elimination plug-ins and Q inhalation artifact elimination plug-ins, the constructed respiratory artifact elimination model can carefully process artifacts in different respiratory stages, significantly improve the quality of chest CT images, and make CT scan images more accurate and reliable.

[0049] S40: Using the Q exhalation artifact elimination plug-ins and the Q inhalation artifact elimination plug-ins, respectively perform respiratory artifact mapping elimination on the 2Q local CT image set to obtain a 2Q standard local CT image set, and after fusion, obtain a reconstructed chest CT image as an image processing result.

[0050] Further, step S40 of the present invention further includes: S41: randomly selecting a first local CT image set of the first exhalation stage or the first inhalation stage from the 2Q local CT image set, and matching to obtain a first artifact removal plug-in, wherein the first artifact removal plug-in is a first exhalation artifact removal plug-in or a first inhalation artifact removal plug-in.

[0051] Specifically, first, any local CT image set of the first exhalation stage or the first inhalation stage is randomly selected from the 2Q local CT image set and set as the first local CT image set. After the first local CT image set is selected, the corresponding artifact removal plug-in is matched according to the respiratory stage information of the image set. If the selected image set belongs to the first exhalation stage in the respiratory cycle, the first exhalation artifact removal plug-in is used to remove artifacts. This plug-in is specifically targeted at the exhalation stage of respiratory artifacts and helps to remove artifacts caused by chest expansion. If the selected image set belongs to the first inhalation stage of the inhalation stage, the first inhalation artifact removal plug-in is used. This plug-in targets the artifact problem caused by chest contraction and helps to eliminate artifacts generated during inhalation.

[0052] S42: Setting the number P of adaptation branch selections based on the user's weight index, where P is less than or equal to K.

[0053] Further, step S42 of the present invention further includes: S421: Obtain the user's body weight index, and calculate the deviation ratio between the user's body weight index and the preset standard body weight index to obtain an individual deviation coefficient; S422: Multiply the sum of 1 and the individual deviation coefficient by a predetermined number of selections and round it to obtain P, where the predetermined number of selections is 5.

[0054] Specifically, first, obtain the user's body mass index, for example, by measuring the user's weight (kg) and height (m) to calculate the user's body mass index; then calculate the deviation ratio between the user's body mass index and a preset standard body mass index. The standard body mass index is usually a fixed value, which can be set to, for example, 22 (the standard value of normal weight), or set according to actual conditions. The deviation ratio is the ratio of the absolute value of the difference between the user's body mass index and the preset standard body mass index to the standard body mass index. The deviation ratio describes the degree of deviation of the user's body mass index from the standard body mass index, and the deviation ratio is set as the individual deviation coefficient.

[0055] Next, the sum of 1 and the individual deviation coefficient is multiplied by the predetermined number of selections (5) and rounded to obtain the number of adaptation branches selected P. For example, assuming that the individual deviation coefficient is 0.2, P is 1.2*5 rounded to 6; the calculated P value will determine the number of adaptation branches selected during the processing. The purpose of this step is to dynamically adjust the number of selected processing branches according to the deviation between the individual's body mass index and the standard body mass index, so as to better adapt to the individual differences of different users. For example, if the individual deviation coefficient is large (i.e., the weight deviates greatly from the standard), more adaptation branches are selected to ensure better processing of individual differences; if the individual deviation coefficient is small (i.e., the weight is close to the standard), fewer adaptation branches are selected to improve processing efficiency.

[0056] By dynamically calculating the number of adaptation branches to be selected based on the deviation between the user's body mass index and the standard value, image processing can be optimized based on individual body shape differences, thereby improving image quality while ensuring processing efficiency.

[0057] S43: randomly selecting P artifact removal branches from the K artifact removal branches of the first artifact removal plug-in, performing artifact removal on the first local CT image set, outputting a first standard local CT image set after mean calculation, and adding it to the 2Q standard local CT image set.

[0058] Specifically, P artifact removal branches are randomly selected from the K artifact removal branches of the first artifact removal plug-in, and artifact removal is performed on the first local CT image set. By randomly selecting P branches, the diversity and robustness of the training process can be increased, and overfitting of a single model can be avoided. After the P branches have processed the first local CT image set respectively, the processed image results are averaged. The mean calculation means averaging the P artifact-eliminated images to obtain a standard output image. This output image represents the collective judgment of all P branches, which can effectively reduce the deviation caused by the error of a single branch, provide a more accurate artifact removal effect, and obtain the first standard local CT image set, which is added to the 2Q standard local CT image set.

[0059] By randomly selecting P from K artifact removal branches for artifact removal and fusing the processing results of multiple branches through mean calculation, a more accurate first standard local CT image set is obtained. This process improves the accuracy of artifact removal and ensures the quality of the image.

[0060] Then, the 2Q standard local CT image set is fused, that is, each local CT image is spliced ​​along the scanning layer direction to gradually construct a complete three-dimensional chest CT image. A seamless overall image can be obtained by splicing, that is, the reconstructed chest CT image, and the reconstructed chest CT image is used as the image processing result.

[0061] A chest CT image processing method provided by an embodiment of the present invention has at least the following technical effects: By obtaining the chest CT image to be processed, as well as the layer-by-layer scanning record and respiratory monitoring record during CT scanning; then based on the respiratory monitoring record, the scanning time zone is divided into Q breathing time zones and Q breathing time zones, and the chest CT image to be processed is divided, clustered and marked according to the Q breathing time zones and Q breathing time zones to obtain a 2Q local CT image set; then a respiratory artifact elimination model is constructed based on a convolutional neural network, wherein the respiratory artifact elimination model includes Q breathing artifact elimination plug-ins and Q breathing artifact elimination plug-ins; finally, the Q breathing artifact elimination plug-ins and Q breathing artifact elimination plug-ins are used to respectively perform respiratory artifact mapping elimination on the 2Q local CT image set to obtain a 2Q standard local CT image set, and after fusion, a reconstructed chest CT image is obtained as an image processing result. In other words, by dividing the image into multiple breathing time zones and using a deep learning model to eliminate artifacts, the precision and accuracy of respiratory artifact elimination can be significantly improved, thereby effectively improving the quality and accuracy of chest CT images.

[0062] Although preferred embodiments of the present invention have been described, additional changes and modifications may occur to these embodiments once those skilled in the art understand the basic inventive concepts.

[0063] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of the present invention and its equivalent technologies, the present invention is also intended to include these changes and variations.

Claims

1. A chest CT image processing method, characterized in that: Methods include: Obtain chest CT images to be processed, as well as layer-by-layer motion scanning records and respiratory monitoring records during CT scanning; Based on the respiratory monitoring record, the scanning time zone is divided into Q exhalation time zones and Q inhalation time zones, and the chest CT image to be processed is divided, clustered and marked according to the Q exhalation time zones and the Q inhalation time zones to obtain a 2Q local CT image set; Constructing a breathing artifact elimination model based on a convolutional neural network, wherein the breathing artifact elimination model includes Q exhalation artifact elimination plug-ins and Q inhalation artifact elimination plug-ins; The Q exhalation artifact elimination plug-ins and the Q inhalation artifact elimination plug-ins are used to respectively perform respiratory artifact mapping elimination on the 2Q local CT image set to obtain a 2Q standard local CT image set, and after fusion, a reconstructed chest CT image is obtained as an image processing result.

2. A chest CT image processing method according to claim 1, characterized in that: Based on the respiratory monitoring record, the scanning time zone is divided into Q exhalation time zones and Q inhalation time zones, including: When the user undergoes a chest CT scan, the chest strap sensor synchronously collects the user's breathing characteristics according to a predetermined time stamp to obtain a breathing feature sequence as a breathing monitoring record; Based on the respiratory monitoring record, the scanning time zone is divided according to Q exhalation phases and Q inhalation phases to obtain Q exhalation time zones and Q inhalation time zones.

3. A chest CT image processing method according to claim 2, characterized in that: According to the chest fluctuation ratio, the user's breathing cycle is subdivided into Q exhalation phases and Q inhalation phases, where Q is an integer greater than 2, the chest expansion ratio of each exhalation phase is different, and the chest contraction ratio of each inhalation phase is different.

4. A chest CT image processing method according to claim 2, characterized in that: The chest CT image to be processed is divided, clustered and marked according to the Q exhalation time zones and the Q inhalation time zones to obtain a 2Q local CT image set, including: Aligning the respiratory monitoring record and the layer-by-layer motion scanning record in the time domain, dividing the chest CT image to be processed according to the Q exhalation time zones and the Q inhalation time zones, and clustering the local CT images in the same exhalation time zone or the same inhalation time zone; The clustering results are stage-labeled according to the Q exhalation stages and the Q inhalation stages, and Q exhalation stage local CT image sets and Q inhalation stage local CT image sets are obtained, which are combined to obtain 2Q local CT image sets.

5. A chest CT image processing method according to claim 4, characterized in that: A respiratory artifact elimination model is built based on a convolutional neural network, including: Randomly select the first breathing stage from the Q breathing stages, and obtain the first chest expansion ratio of the first breathing stage; Taking the first chest expansion ratio as a feature constraint, retrieving a first training set to perform supervised training on a convolutional neural network, and obtaining a first call artifact elimination plug-in; Based on the first exhalation artifact elimination plug-in, the Q exhalation stages and the Q inhalation stages are analyzed respectively to obtain Q exhalation artifact elimination plug-ins and Q inhalation artifact elimination plug-ins, which are combined to construct a respiratory artifact elimination model.

6. A chest CT image processing method according to claim 5, characterized in that: Taking the first chest expansion ratio as a feature constraint, retrieving a first training set to perform supervised training on a convolutional neural network, and obtaining a first call artifact elimination plug-in, including: Taking the first chest expansion ratio as a feature constraint, searching chest CT scan logs in a historical time zone, collecting sample first local CT images of multiple users, and obtaining sample first standard local CT images of multiple users, to obtain a sample first local CT image set and a sample first standard local CT image set, wherein the standard local CT image is a chest CT image without respiratory interference; The sample first local CT image set and the sample first standard local CT image set are used as sample data sets, and are equally divided into K parts to obtain K training sets, where K is an integer greater than 10; The K training sets are used to perform supervised training on the convolutional neural network respectively until the model converges, and K first call artifact elimination branches are obtained, which are integrated to construct the first call artifact elimination plug-in.

7. A chest CT image processing method according to claim 6, characterized in that: Respiratory artifact mapping is performed on the 2Q local CT image sets respectively to obtain a 2Q standard local CT image set, including: Randomly selecting a first local CT image set of the first exhalation stage or the first inhalation stage from the 2Q local CT image set, and matching to obtain a first artifact elimination plug-in, wherein the first artifact elimination plug-in is a first exhalation artifact elimination plug-in or a first inhalation artifact elimination plug-in; The number of adaptation branch selections P is set based on the user's weight index, where P is less than or equal to K; P artifact removal branches are randomly selected from the K artifact removal branches of the first artifact removal plug-in, artifact removal is performed on the first local CT image set, and a first standard local CT image set is output after mean calculation, and added to the 2Q standard local CT image set.

8. A chest CT image processing method according to claim 7, characterized in that: The number P of adaptation branch selection is set based on the user's weight index, including: Obtaining a user's body mass index, and calculating a deviation ratio between the user's body mass index and a preset standard body mass index to obtain an individual deviation coefficient; Multiply the sum of 1 and the individual deviation coefficient by the predetermined number of selections and round it to the integer to obtain P, where the predetermined number of selections is 5.

Images