CT sequence scanning range efficient classification method based on maximum density projection

Abstract

The invention provides a CT sequence scanning range efficient classification method based on maximum density projection, and relates to the field of medical image processing. Reducing the dimension of a three-dimensional CT sequence image in the target CT sequence data into a two-dimensional MIP image by using maximum density projection; constructing an image classification model, inputting the two-dimensional MIP image into the image classification model for classification training, calculating classification loss through labels of a training set, and performing parameter optimization; and after the training of the image classification model is completed, inputting unlabeled target CT sequence data, converting the unlabeled target CT sequence data into a two-dimensional MIP image by using maximum density projection, and predicting and classifying a scanning range by the image classification model. According to the method, the computing resource and memory requirements required by CT image processing can be reduced, and the efficiency of CT image data processing is remarkably improved.

Description

Technical Field

[0001] The invention relates to the field of medical image processing, and in particular to an efficient classification method for CT sequence scanning range based on maximum intensity projection (MIP). Background Art

[0002] With the development of medical imaging technology, computed tomography (CT) has become one of the indispensable tools in medical diagnosis. However, with the rapid growth of CT image data, how to effectively process and classify this data has become an important technical issue. Especially in the field of medical deep learning, different tasks usually focus on different parts of the patient's body, so the accurate distinction of the scanning range of CT data is particularly critical in the data cleaning process. However, the existing method of distinguishing the CT scanning range mainly relies on manual recognition and classification by medical experts, which has the following significant limitations:

[0003] (1) High labor cost: Manual identification requires medical experts to review the CT series one by one and judge the scanning range based on the anatomical structure of different parts. Although this method is accurate, it is extremely time-consuming and labor-intensive, and is very inefficient, especially when processing large-scale data sets.

[0004] (2) Dependence on professional technicians: The recognition and classification of CT images requires personnel with professional medical knowledge. This not only increases the cost of data processing, but also in some resource-limited areas, the insufficient supply of medical experts further limits the efficiency of data processing.

[0005] In order to improve the automation of CT scan range distinction, image classification methods based on deep learning have been widely used. However, these traditional deep learning methods also have the following problems when applied to CT images:

[0006] (1) Complex data annotation: CT images are usually three-dimensional data, and each three-dimensional image contains a large number of two-dimensional tomographic images. During the data annotation process, medical experts need to find key images from these tomographic images and manually annotate them, which is not only time-consuming but also prone to human errors.

[0007] (2) High computing resource requirements: Due to the three-dimensional characteristics of CT images, traditional methods usually require the input of the entire CT sequence for processing, which places high demands on computing resources and video memory. Especially when processing large-scale CT data, the model training process requires a lot of computing time and storage space, resulting in low processing efficiency and high cost. Summary of the invention

[0008] In view of this, in order to solve the above problems, the purpose of the present invention is to propose an efficient classification method for CT sequence scanning range based on Maximum Intensity Projection (MIP), which uses Maximum Intensity Projection (MIP) technology to reduce the dimension of three-dimensional images to two-dimensional images, so as to solve the problems of long training time and high computing resource requirements faced by the prior art in processing CT images, and reduce the computing resources and memory requirements required for processing CT images. Compared with the original three-dimensional CT sequence data, the present invention effectively reduces the amount of data to be processed, reduces the demand for computing resources and memory, can significantly improve the efficiency of CT image data processing, and reduce the cost and resource consumption of data processing, providing a more efficient solution for the cleaning and analysis of large-scale medical data.

[0009] Based on the above purpose, in a first aspect, the present invention provides an efficient classification method for CT sequence scanning range based on maximum density projection, the method comprising the following steps:

[0010] Input target CT sequence data;

[0011] Use Maximum Intensity Projection (MIP) to reduce the dimensionality of the three-dimensional CT sequence images in the target CT sequence data into two-dimensional MIP images;

[0012] Construct an image classification model, input the two-dimensional MIP image into the image classification model for classification training, calculate the classification loss through the labels of the training set and perform parameter optimization;

[0013] After the image classification model is trained, the unlabeled target CT sequence data is input and converted into a two-dimensional MIP image using maximum intensity projection. The image classification model then predicts and classifies the scanning range.

[0014] As a further solution of the present invention, after inputting the target CT sequence data, the method further includes:

[0015] Read the metadata information of the DICOM file in the target CT sequence data, and convert the stored value of each point into Hounsfield units through the RS and RIA values ​​in the metadata, that is:

[0016] HU=RS×SV+RIA

[0017] Wherein, HU stands for Hounsfield unit; SV stands for storage value.

[0018] As a further solution of the present invention, when performing maximum density projection, a corresponding two-dimensional MIP image is generated by selecting a maximum CT value along a specific direction of a target CT sequence.

[0019] As a further solution of the present invention, in the maximum density projection process, the projection direction of the maximum density projection is perpendicular to the coronal plane. When the projection is generated, the maximum density projection calculation is performed on each image layer. The maximum density projection is completed by calculating the maximum value of the CT value on each projection line. The maximum density projection calculation formula is:

[0020] MIP(y,z)=Max(HU(x i ,y,z))

[0021] Wherein, the x-axis is the sagittal axis of the CT sequence image; the y-axis is the coronal axis; the z-axis is the transverse axis; MIP(y,z) is the CT value of the pixel with coordinates (y,z) in the two-dimensional image after the maximum density projection perpendicular to the coronal plane; Max(HU(x i ,y,z)) is the maximum value of the CT values ​​of n pixels along the x direction at the position with coronal axis coordinate y in the zth layer in the CT sequence.

[0022] As a further solution of the present invention, when projecting in a direction perpendicular to the coronal plane, an initial projection point is selected from the three-dimensional image data, and the projection is extended in a direction perpendicular to the coronal plane, and the maximum CT value on each section image passed is recorded; the maximum CT value recorded on each section image is plotted at a corresponding position of the corresponding two-dimensional projection image, and the maximum CT value projection is performed; the projected CT value is windowed, and all calculated projection values ​​are converted into 16-bit grayscale values ​​and a square two-dimensional MIP image is formed.

[0023] As a further solution of the present invention, when the maximum CT value is projected, if a pixel passes through multiple slice images, the maximum CT value among them is selected for projection.

[0024] As a further solution of the present invention, the image classification model is a VGG11 model based on a convolutional neural network (CNN), which includes 11 convolutional layers and 3 fully connected layers, and is used to automatically extract features in two-dimensional MIP images and classify the CT scanning range.

[0025] As a further solution of the present invention, when training the image classification model, a cross-entropy loss function is used to calculate the classification loss, and an optimization algorithm is used to adjust the model parameters. The optimization algorithm includes but is not limited to a stochastic gradient descent (SGD) or an Adam optimization algorithm.

[0026] As a further solution of the present invention, during the model training process, the constructed image classification model is trained using a labeled data set, and the model parameters are iteratively adjusted through a cross-entropy loss function until the model converges on the training set and reaches the expected classification accuracy on the validation set; after the image classification model training is completed, the unlabeled target CT sequence is converted into a two-dimensional MIP image, and the image is input into the trained image classification model, and the model automatically predicts and outputs the classification result of the CT scanning range.

[0027] As a further solution of the present invention, during the prediction process, the image classification model calculates the classification probability of each type of scanning range and selects the category with the highest probability as the final CT scanning range classification result.

[0028] The efficient classification method of CT sequence scanning range based on maximum density projection of the present invention can be applied to the classification of CT scanning ranges of different anatomical parts, including but not limited to the classification of parts such as the head, chest, abdomen, etc. The method is suitable for CT scanning range classification tasks in real-time or batch processing, and can effectively improve the efficiency and accuracy of CT scanning range classification.

[0029] In another aspect of the present invention, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, any one of the above-mentioned efficient classification methods for CT sequence scanning range based on maximum density projection according to the present invention is executed.

[0030] In yet another aspect of the present invention, a computer-readable storage medium is provided, storing computer program instructions, which, when executed, implement any of the above-mentioned methods for efficiently classifying CT sequence scanning ranges based on maximum density projection according to the present invention.

[0031] Compared with the prior art, the present invention proposes an efficient classification method for CT sequence scanning range based on maximum density projection, which has the following beneficial effects:

[0032] 1. Efficient classification and rapid diagnosis are achieved. The present invention uses maximum intensity projection (MIP) technology to reduce the dimensionality of complex three-dimensional CT sequence images to two-dimensional images, greatly simplifying the complexity of data processing and reducing the demand for computing resources. At the same time, MIP technology retains the most representative high-density areas in CT images, allowing key anatomical structures to be highlighted, helping medical experts to quickly identify and annotate the specific range of CT scans (such as head, chest, abdomen, etc.), thereby improving the efficiency and accuracy of clinical diagnosis.

[0033] 2. Significantly improved image quality and recognition accuracy. By selecting the maximum CT value during the projection process and combining the windowing technology to enhance the display effect of the projected image, the two-dimensional MIP image generated by the present invention has significantly improved clarity and contrast. This makes the key structures in the image more obvious, especially the edge features of high-density tissues such as bones, thereby improving the accuracy of the subsequent classification model in identifying the scanning range.

[0034] 3. Simplified model training and improved model performance. Since the MIP technology in the present invention can significantly reduce the dimension of three-dimensional image data, the amount of two-dimensional image data generated is greatly reduced, thereby reducing the training time and computational overhead of the model. The convolutional neural network (CNN) is trained and classified based on two-dimensional MIP images, and its processing speed is improved. While maintaining a high classification accuracy, it effectively reduces the risk of overfitting and improves the generalization ability of the model.

[0035] 4. Improved model robustness and adaptability. Through preprocessing operations such as data enhancement and image normalization, the present invention further improves the robustness of the model, so that it can still maintain a high classification accuracy under different scanning conditions and individual differences among patients. In this way, the recognition results of the CT scan range can be highly reliable for both normal patients and patients with complex diseases.

[0036] 5. Strong scalability and wide applicability. The method of the present invention is not limited to the CT scanning range identification of specific parts. Its flexible model design and efficient projection algorithm can easily adapt to CT scanning data of different parts and types. This means that the method can be widely used in various CT imaging diagnosis scenarios, such as whole-body CT scanning, local enhanced CT scanning, etc., providing a general and efficient solution for clinical practice. The final result of the present invention: the image classification model based on CT scanning range identification based on maximum density projection can be embedded in the actual clinical image analysis system to realize automatic scanning range classification. This not only greatly reduces the time of manual labeling, but also helps clinicians quickly obtain accurate diagnostic information, especially in dealing with a large number of patients or emergency situations, it can effectively improve the quality and efficiency of medical services.

[0037] 6. Support large-scale data processing: Since the present invention effectively reduces the dimensionality of three-dimensional CT images and makes full use of data enhancement technology in the training stage, this method can perform well when processing large-scale CT data. At the same time, it can also adapt to data acquired under different devices and scanning parameters, and has strong versatility and adaptability. Through the efficient classification method of CT scanning range of the present invention, medical institutions can more accurately analyze the condition of patients and formulate personalized treatment plans. This automated and efficient classification technology will help promote the development of precision medicine and further improve the scientificity and rationality of clinical decision-making.

[0038] These and other aspects of the present application will be more concise and understandable in the following description of the embodiments. It should be understood that the above general description and the following detailed description are only exemplary and explanatory and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] 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 ordinary technicians in this field, other embodiments can be obtained based on these drawings without paying creative work.

[0040] In the figure:

[0041] Figure 1 The flowchart is a method for efficiently classifying the scanning range of a CT sequence based on maximum density projection according to an embodiment of the present invention.

[0042] Figure 2 It is a principle diagram of maximum density projection in the efficient classification method of CT sequence scanning range based on maximum density projection according to an embodiment of the present invention.

[0043] Figure 3 It is a schematic diagram of the image features of the head bone tissue highlighted in the two-dimensional MIP image in the efficient classification method of the CT sequence scanning range based on maximum density projection in an embodiment of the present invention.

[0044] Figure 4 It is a schematic diagram of the image features of the abdominal bone tissue highlighted in the two-dimensional MIP image in the efficient classification method of the CT sequence scanning range based on maximum density projection in an embodiment of the present invention.

[0045] Figure 5 It is a schematic diagram of the image features of chest bone tissue highlighted in a two-dimensional MIP image in an efficient classification method for CT sequence scanning range based on maximum density projection according to an embodiment of the present invention.

[0046] Figure 6The present invention is a schematic diagram of the workflow of the automatic classification service of CT scanning range in the efficient classification method of CT sequence scanning range based on maximum density projection in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] Below, the present application is further described in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form a new embodiment.

[0048] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in combination with specific embodiments and with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0049] The technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] Since manual differentiation of CT image data has the disadvantages of consuming manpower and material resources and requiring professional technicians, the traditional image classification method based on deep learning also has the disadvantages of high data labeling requirements, high resource requirements, and long time consumption. In order to overcome the shortcomings of the prior art in identifying the CT scanning range, the present invention provides an efficient classification method for the CT sequence scanning range based on maximum density projection, which reduces the dimension of the three-dimensional image to a two-dimensional image by using the maximum density projection (Maximum Intensity Projection, MIP) technology to solve the problems of long training time and high computing resource requirements faced by the prior art in processing CT images, and reduce the computing resources and memory requirements required for processing CT images. Compared with the original three-dimensional CT sequence data, the amount of data to be processed and the complexity of the algorithm are greatly reduced. This means that the training time can be significantly shortened, while also reducing the requirements for hardware equipment and improving overall efficiency.

[0051] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0052] See also Figure 1 As shown, an embodiment of the present invention provides an efficient classification method for CT sequence scanning range based on maximum density projection, and the method includes the following steps:

[0053] Step S10: input target CT sequence data.

[0054] Step S20, using maximum intensity projection to reduce the dimension of the three-dimensional CT sequence images in the target CT sequence data into two-dimensional MIP images;

[0055] Step S30, constructing an image classification model, inputting the two-dimensional MIP image into the image classification model for classification training, calculating the classification loss through the labels of the training set and performing parameter optimization;

[0056] Step S40: After the image classification model is trained, unlabeled target CT sequence data is input, and the unlabeled target CT sequence data is converted into a two-dimensional MIP image using maximum intensity projection, and the image classification model is used to predict and classify the scanning range.

[0057] In the present invention, a maximum intensity projection (MIP) method is first proposed to reduce the dimension of three-dimensional CT images to two-dimensional images. Secondly, in order to maximize the low-dimensional advantage of two-dimensional images, the present invention uses convolutional neural networks (CNN) VGG11 (Visual Geometry Group 11-layer model) to build a prediction model. Through the above method, efficient automatic distinction of CT scanning range is achieved.

[0058] After the target CT sequence data is input in step S10, the following further includes: reading metadata information of the DICOM file in the target CT sequence data, and converting the stored value of each point into Hounsfield units according to RS and RIA values ​​in the metadata, that is, HU = RS × SV + RIA

[0059] Wherein, HU stands for Hounsfield unit; SV stands for storage value.

[0060] The present invention collects a large amount of CT sequence data and converts these three-dimensional images into two-dimensional images through MIP technology. MIP (Maximum Intensity Projection) is used to convert three-dimensional images into two-dimensional projection images and highlight the maximum CT value in the image. Among them, when MIP is processed, it includes data acquisition and preprocessing, conversion process, projection generation, projection along the direction perpendicular to the coronal plane, maximum CT value projection and synthetic projection.

[0061] Among them, when collecting data and reducing the dimension of projection, before applying the maximum density projection, the CT sequence data is first obtained, the original three-dimensional CT data is preprocessed, and the metadata information of the DICOM (Digital Imaging and Communications in Medicine) file of the CT image is first read, and the stored values ​​(SV) of each point are converted into Hounsfield Units (HU) through the Rescale Slope (RS) and Rescale Intercept Attribute (RIA) values ​​in the metadata.

[0062] Then, during the conversion process, the corresponding two-dimensional MIP image is generated by selecting the maximum CT value along a specific direction of the target CT sequence.

[0063] When generating projections, the projection direction of the maximum density projection is perpendicular to the coronal plane. When generating projections, the maximum density projection calculation is performed for each image layer. The maximum density projection is completed by calculating the maximum value of the CT value on each projection line, such as Figure 2 As shown, the maximum density projection calculation formula is:

[0064] MUP(y,z)=Max(HU(x i ,y,z))

[0065] Wherein, the x-axis is the sagittal axis of the CT sequence image; the y-axis is the coronal axis; the z-axis is the transverse axis; MIP(y,z) is the CT value of the pixel with coordinates (y,z) in the two-dimensional image after the maximum density projection perpendicular to the coronal plane; Max(HU(x i ,y,z)) is the maximum value of the CT values ​​of n pixels along the x direction at the position with coronal axis coordinate y in the zth layer in the CT sequence.

[0066] When projecting in a direction perpendicular to the coronal plane, an initial projection point is selected from the three-dimensional image data, and the projection is extended in a direction perpendicular to the coronal plane, and the maximum CT value on each section image passed is recorded; the maximum CT value recorded on each section image is plotted at the corresponding position of the corresponding two-dimensional projection image, and the maximum CT value projection is performed; finally, when synthesizing the projection, the projected CT value is windowed to enhance its display clarity, and all calculated projection values ​​are converted into 16-bit grayscale values ​​to form a square two-dimensional MIP image.

[0067] When the maximum CT value is projected, if a pixel passes through multiple slice images, the maximum CT value is selected for projection.

[0068] like Figure 3-Figure 5As shown in the figure, through maximum density projection, the MIP images of the three parts highlight the imaging characteristics of the bone tissue, so the difference in the scanning range can be very clearly seen, and the most easily distinguished bone density map is retained to the maximum extent. Therefore, the main advantage of MIP is to maximize the retention of high-density features of body parts, and can highlight the greatest characteristics of the structure in the body, which helps to quickly observe and analyze important scanning ranges in the image and highlight key information.

[0069] In the embodiment of the present invention, after the three-dimensional CT sequence images in the target CT sequence data are reduced to two-dimensional MIP images in step S20, data annotation is also included, and medical experts accurately and quickly annotate each two-dimensional MIP image to indicate the scanning range of the CT sequence. For example: head, chest, abdomen, etc. The collected two-dimensional MIP image data is preprocessed, including data enhancement and image normalization operations. These preprocessing steps help improve the accuracy and robustness of the model.

[0070] In an embodiment of the present invention, in step S30, the image classification model is a VGG11 model based on a convolutional neural network (CNN), and the VGG11 model includes 11 convolutional layers and 3 fully connected layers, which are used to automatically extract features in two-dimensional MIP images and classify the CT scanning range.

[0071] Among them, the VGG11 model has a deeper network structure and can obtain more feature information. The VGG11 model consists of 11 convolutional layers and 3 fully connected layers, with a total of about 140M parameters. During the training process, the network input of the present invention is a fixed-size 224×224 image. The image passes through a series of convolutional layers, in which the convolution kernel size is 3×3 and the convolution stride is fixed to 1. In order to maintain the original spatial resolution, a padding operation is performed after the convolution, that is, for the 3×3 convolutional layer, the padding number is 1 pixel. In the network, five spatial maximum pooling layers are alternated with the convolutional layers, the pooling window is 2×2, and the pooling stride is 2. All spatial downsampling is completed through the maximum pooling layer. After a series of convolution and pooling layer processing, a flattened feature vector is obtained, which is input into three fully connected layers, the first two layers are 4096 dimensions, and the last layer dimension is the number of types in the scanning range. The first fully connected layer is randomly inactivated with a probability of 50%, and the output of the last layer is the classification score of a given image. The class corresponding to the maximum score is the classification result. Compared with other similar neural network models, VGG11 adopts a very simple design in network structure, including only small-sized convolution kernels, pooling layers, and ordinary fully connected layers, and only uses ReLU (Rectified Linear Unit) activation function for nonlinear transformation. This simple and practical architecture is easy to train to extract features from images and make predictions to distinguish scanning ranges.

[0072] When training the image classification model, a cross-entropy loss function is used to calculate the classification loss, and an optimization algorithm is used to adjust the model parameters. The optimization algorithm includes but is not limited to a stochastic gradient descent (SGD) or an Adam optimization algorithm.

[0073] During the model training and tuning process, the constructed image classification model is trained using a labeled data set, and the model parameters are iteratively adjusted through a cross-entropy loss function until the model converges on the training set and reaches the expected classification accuracy on the validation set. After the image classification model training is completed, the unlabeled target CT sequence is converted into a two-dimensional MIP image and input into the trained image classification model, and the model automatically predicts and outputs the classification results of the CT scanning range.

[0074] Among them, when the loss function adopts the cross entropy loss function, the formula of the cross entropy loss function is:

[0075]

[0076] In the formula, p iis the i-th output of the classifier, y i is the true label corresponding to the i-th output.

[0077] The model is iteratively trained on the training set, and the model parameters are continuously adjusted through the back-propagation algorithm until the convergence condition is reached, so that the model can better learn the image features and distinguish the scanning range.

[0078] In this embodiment, an independent test set is used to evaluate and verify the trained model during model evaluation and verification. The accuracy, recall, and precision indicators of the model on the test set are calculated. The classification effect of the model in different scanning ranges is analyzed, whether the model can correctly identify each category is checked, and the performance and generalization ability of the model are evaluated. If the model performs poorly, the model can be further optimized by adjusting the model structure or hyperparameters. Adjusting the model structure generally includes adding new convolutional layers, introducing residual connections and attention mechanisms, and other operations. Use model ensemble technology, that is, training multiple models, and executing all models simultaneously during classification. The final classification score is the average of the classification scores of all models, so as to improve the performance and generalization ability of the model.

[0079] When the model is deployed and applied, the model is exported, embedded, and tested. When exporting the model, the trained model is exported to a deployable format, such as TensorFlow SavedModel or ONNX model. When embedding the model, the exported model is embedded into the production environment in the form of a service or a library. When testing the model, the model is tested in the production environment to ensure that it works properly and the output results meet expectations. This model can be deployed as a web service, packaged into a docker image, or deployed on an embedded device to automatically classify and identify the scan range and realize an automated data cleaning process.

[0080] For example, a typical CT scan range automatic classification service workflow is as follows: Figure 6 As shown in the figure, for the input CT sequence, a set of classification scores is obtained through MIP dimension reduction and VGG network inference. The class corresponding to the largest score is the classification result. The service automatically outputs the scanning range of the CT sequence.

[0081] In this embodiment, during the prediction process, the image classification model calculates the classification probability of each type of scanning range and selects the category with the highest probability as the final CT scanning range classification result.

[0082] The efficient classification method of CT sequence scanning range based on maximum density projection of the present invention can be applied to the classification of CT scanning ranges of different anatomical parts, including but not limited to the classification of parts such as the head, chest, abdomen, etc. The method is suitable for CT scanning range classification tasks in real-time or batch processing, and can effectively improve the efficiency and accuracy of CT scanning range classification.

[0083] In summary, the present invention's efficient classification method for CT sequence scanning range based on maximum density projection includes the steps of data collection and MIP dimensionality reduction projection, data preprocessing, model construction and training, model training and tuning, model evaluation and verification, deployment and application. Through these steps, the scanning range distinction of CT images based on the MIP method can be achieved, and the accuracy and efficiency of the scanning range distinction data cleaning process can be improved. The present invention uses maximum density projection to convert a three-dimensional CT sequence into a two-dimensional image: MIP technology generates an image by selecting the maximum CT value, while retaining the key features of scanning range distinction, the three-dimensional image projection is compressed into a two-dimensional image. Traditional three-dimensional image technology usually uses the original three-dimensional sequence image to distinguish the scanning range, resulting in larger data, slow operation speed, and occupying more space, which makes MIP image technology have advantages in information storage and transmission operations.

[0084] The present invention also utilizes an artificial intelligence automated process to replace manual scanning range differentiation: this method automatically differentiates the scanning range by using a deep learning model. Compared with manual differentiation, the differentiation speed is faster and more efficient, and it does not require the time of doctors and experts, thus saving a lot of manpower and material resources.

[0085] Therefore, the MIP technology used in the present invention is different from the traditional three-dimensional image technology in terms of specific technical means. The MIP technology is simple and has a fast calculation speed, low machine requirements, and the model can be reused after it is built, without consuming too much manpower and material resources.

[0086] According to a second aspect of an embodiment of the present invention, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method of any one of the above embodiments is implemented.

[0087] The computer device includes a processor and a memory, and may also include an input system and an output system. The processor, memory, input system and output system may be connected via a bus or other means, and the input system may receive input digital or character information, and generate signal input related to the migration of the efficient classification method of the CT sequence scanning range based on maximum density projection. The output system may include display devices such as a display screen.

[0088] As a non-volatile computer-readable storage medium, the memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as the program instructions / modules corresponding to the efficient classification method for the CT sequence scanning range based on maximum density projection in the embodiment of the present application. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the use of the efficient classification method for the CT sequence scanning range based on maximum density projection, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the local module via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0089] The processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor is generally used to control the overall operation of the computer device. In this embodiment, the processor is used to run the program code stored in the memory or process data. The processors of multiple computer devices of the computer device of this embodiment execute various functional applications and data processing of the server by running non-volatile software programs, instructions and modules stored in the memory, that is, implementing the steps of the efficient classification method of the CT sequence scanning range based on maximum density projection of the above method embodiment.

[0090] It should be understood that, to the extent that they do not conflict with each other, all the embodiments, features and advantages described above for the efficient classification method for the scanning range of a CT sequence based on maximum density projection according to the present invention are also applicable to the efficient classification method for the scanning range of a CT sequence based on maximum density projection and the storage medium according to the present invention.

[0091] It will also be appreciated by those skilled in the art that various exemplary logic blocks, modules, circuits and algorithm steps described in conjunction with the disclosure herein can be implemented as electronic hardware, computer software or a combination of the two. In order to clearly illustrate this interchangeability of hardware and software, a general description has been given to the functions of various schematic components, blocks, modules, circuits and steps. Whether this function is implemented as software or hardware depends on specific applications and the design constraints imposed on the entire system. Those skilled in the art can implement the function in various ways for each specific application, but this implementation decision should not be interpreted as causing a departure from the disclosed scope of the embodiments of the present invention.

[0092] The above are exemplary embodiments disclosed in the present invention, but it should be noted that various changes and modifications may be made without departing from the scope disclosed in the embodiments of the present invention as defined in the claims. The functions, steps and / or actions of the method claims according to the disclosed embodiments described herein do not need to be performed in any particular order. In addition, although the elements disclosed in the embodiments of the present invention may be described or required in individual form, they may also be understood as multiple unless explicitly limited to the singular.

[0093] It should be understood that, as used herein, the singular form "a" or "an" is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, "and / or" refers to any and all possible combinations of one or more of the items listed in association. The serial numbers of the embodiments disclosed in the above embodiments of the present invention are for description only and do not represent the advantages and disadvantages of the embodiments.

[0094] A person skilled in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the disclosure of the embodiments of the present invention (including the claims) is limited to these examples; under the concept of the embodiments of the present invention, the technical features in the above embodiments or different embodiments can also be combined, and there are many other changes in different aspects of the embodiments of the present invention as above, which are not provided in detail for the sake of simplicity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the protection scope of the embodiments of the present invention.

Claims

1. An efficient classification method for CT sequence scanning range based on maximum density projection, characterized in that: The following steps are involved: Input target CT sequence data; Use maximum intensity projection to reduce the dimension of the three-dimensional CT sequence images in the target CT sequence data into two-dimensional MIP images; Construct an image classification model, input the two-dimensional MIP image into the image classification model for classification training, calculate the classification loss through the labels of the training set and perform parameter optimization; After the image classification model is trained, the unlabeled target CT sequence data is input and converted into a two-dimensional MIP image using maximum intensity projection. The image classification model then predicts and classifies the scanning range.

2. The method for efficiently classifying the scanning range of a CT sequence based on maximum density projection according to claim 1, characterized in that: After inputting the target CT sequence data, it also includes: Read the metadata information of the DICOM file in the target CT sequence data, and convert the stored value of each point into Hounsfield units through the RS and RIA values ​​in the metadata, that is: HU=RS×SV+RIA Wherein, HU stands for Hounsfield unit; SV stands for storage value.

3. The method for efficiently classifying the scanning range of CT sequences based on maximum density projection according to claim 2, characterized in that: When performing maximum intensity projection, the corresponding two-dimensional MIP image is generated by selecting the maximum CT value along a specific direction of the target CT sequence.

4. The method for efficiently classifying the scanning range of CT sequences based on maximum density projection according to claim 3, characterized in that: In the process of maximum density projection, the projection direction of the maximum density projection is perpendicular to the coronal plane. When the projection is generated, the maximum density projection calculation is performed for each image layer. The maximum density projection is completed by calculating the maximum value of the CT value on each projection line. The maximum density projection calculation formula is: MIP(y,z)=Max(HU(x i ,y,z)) Wherein, the x-axis is the sagittal axis of the CT sequence image; the y-axis is the coronal axis; the z-axis is the transverse axis; MIP(y,z) is the CT value of the pixel with coordinates (y,z) in the two-dimensional image after the maximum density projection perpendicular to the coronal plane; Max(HU(x i ,y,z)) is the maximum value of the CT values ​​of n pixels along the x direction at the position with coronal axis coordinate y in the zth layer in the CT sequence.

5. The method for efficiently classifying the scanning range of CT sequences based on maximum density projection according to claim 4, characterized in that: When projecting in a direction perpendicular to the coronal plane, an initial projection point is selected from the three-dimensional image data, and the projection is extended in a direction perpendicular to the coronal plane, and the maximum CT value on each slice image is recorded; The maximum CT value recorded on each section image is plotted at the corresponding position of the corresponding two-dimensional projection image to perform maximum CT value projection; the projected CT value is windowed, all calculated projection values ​​are converted to 16-bit grayscale values ​​and a square two-dimensional MIP image is formed.

6. The method for efficiently classifying the scanning range of CT sequences based on maximum density projection according to claim 5, characterized in that: When the maximum CT value is projected, if a pixel passes through multiple slice images, the maximum CT value is selected for projection.

7. The efficient classification method for CT sequence scanning range based on maximum density projection according to any one of claims 1 to 6, characterized in that: The image classification model is a VGG11 model based on a convolutional neural network, which includes 11 convolutional layers and 3 fully connected layers, and is used to automatically extract features from two-dimensional MIP images and classify the CT scanning range.

8. The method for efficiently classifying CT sequence scanning range based on maximum density projection according to claim 7, characterized in that: When training the image classification model, the classification loss is calculated using a cross entropy loss function, and the model parameters are adjusted using an optimization algorithm, which includes a stochastic gradient descent or an Adam optimization algorithm.

9. The method for efficient classification of CT sequence scanning range based on maximum density projection according to claim 8, characterized in that: During the model training process, the constructed image classification model is trained using the labeled data set, and the model parameters are iteratively adjusted through the cross entropy loss function until the model converges on the training set and reaches the expected classification accuracy on the validation set. After the image classification model training is completed, the unlabeled target CT sequence is converted into a two-dimensional MIP image and input into the trained image classification model, and the model automatically predicts and outputs the classification results of the CT scanning range.

10. The method for efficient classification of CT sequence scanning range based on maximum density projection according to claim 9, characterized in that: During the prediction process, the image classification model calculates the classification probability of each type of scanning range and selects the category with the highest probability as the final CT scanning range classification result.

Images