Intelligent image processing method and system for acute cerebral hemorrhage CT images

Abstract

The present invention relates to the field of image processing, and discloses an intelligent image processing method and system for acute cerebral hemorrhage CT images. The processing method comprises the following steps: S1: database storage, obtaining original CT image data of acute cerebral hemorrhage in a mobile operating room; S2: data resolution enhancement, to obtain CT images with enhanced resolution; S3: neural network image segmentation, for performing neural network-based image segmentation on CT images with enhanced resolution; S4: reconstruction, for reconstructing the segmented images, reconstructing the patient's head CT data and the segmented lesion site data; S5: establishing visualization and interface, obtaining a head CT image and a three-dimensional model of the bleeding site and a head CT two-dimensional image of different viewing angles; S6: neural network prediction, obtaining a prediction result through improved CNN network training. Aiming at the characteristics of CT images of acute cerebral hemorrhage in a mobile operating room, the present invention performs image preprocessing, segmentation, reconstruction and prediction in steps to adapt to the application of mobile CT.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to an intelligent image processing method and system for acute cerebral hemorrhage CT images. Background Art

[0002] Mobile CT is an important development direction in the field of CT in recent years. The emergence of mobile CT can advance the scene of medical imaging examination. For patients, it can reduce the time of diagnosis, and facilitate the attending physician to make a diagnosis and obtain a treatment plan in the first time. Mobile CT has the characteristics of miniaturization, convenience, low dose, and low energy consumption. It can adapt to various application scenarios such as emergency treatment of stroke, emergency treatment of critical craniocerebral diseases, rapid diagnosis of acute cerebrovascular diseases and pre-hospital craniocerebral trauma; it can be widely used in multiple scenes such as emergency clinics, ICUs, operating rooms and bedsides in hospitals, providing patients with the most timely and accurate diagnosis and rescue. Mobile CT can quickly check through mobile CT and use 5G network to transmit data to the hospital in real time, so as to realize early knowledge of the condition and early preparation for surgery, save the time of examination after admission, and reserve the "golden time" for rescue. On the one hand, the emergence of mobile CT solves the problem of long installation cycle of new equipment and alleviates the problem of insufficient hospital space and equipment. On the other hand, the mobile CT scanning unit can set up an independent inspection unit, which is flexible, effective, and mobile. It can be operated in isolation, realizing zero contact between doctors and patients and reducing the probability of cross infection in the hospital.

[0003] Image processing is performed on CT images of mobile operating rooms. Mobile CT has the characteristics of low dose, short scanning time, few layers, thick layers, low resolution and low contrast. Therefore, it is necessary to use corresponding interpolation algorithms to improve the resolution of the original image, perform interpolation calculations on two-dimensional and three-dimensional images, and improve the image reconstruction effect. In addition, since the number of mobile CT scanning layers is small, only 32 layers, and the number of layers with bleeding sites accounts for less than half of the total number of layers, data distribution is also one of the important reasons affecting the segmentation effect. Therefore, an intermediate layer prediction network is created to increase the number of intermediate layers of CT images, improve data resolution, and divide the data set into normal data and bleeding data, and perform data enhancement on the bleeding data to keep the data volume of bleeding data and normal data balanced.

[0004] The defects of the prior art are summarized as follows:

[0005] Defect 1: Some existing clinical medical image processing systems have relatively simple functions and single processing objects, which cannot meet the growing clinical needs in my country. Most image processing systems only support image browsing functions, and other functions such as image segmentation and 3D reconstruction require the assistance of other systems. The process is cumbersome and the learning cost for clinical staff is high, which is not conducive to promotion and use.

[0006] For example, Horos is an open source medical image browsing software for Mac operating systems. It allows users to load and view medical images in DICOM format, and provides basic image browsing, zooming, window width and window position adjustment functions. However, its functions are relatively simple and may not be suitable for scenarios that require complex analysis or processing.

[0007] For example, MicroDicom Viewer is a free medical image browsing software for Windows operating system. It supports multiple image formats, including DICOM, and provides basic image browsing, zoom in, zoom out, flip and other functions. Due to the limitation of functions, it may not meet the needs of users who need more functions.

[0008] For example, 3D Slicer is an open source medical image processing software that is specifically used to analyze, visualize and process medical image data, including CT, MRI, etc. It provides a wealth of functions and tools for medical research, clinical diagnosis and surgical planning. Problems: Since 3D Slicer has a wealth of functions and tools, beginners may need some time to learn how to use the software; the image processing results of 3D Slicer depend on the image quality. If the original image resolution is insufficient, it may affect the accuracy of subsequent analysis and processing; although 3D Slicer can segment the lesion site, it is still a semi-automatic segmentation based on thresholds, which requires a lot of time and effort, and the segmentation accuracy is more dependent on the physician's subjective judgment.

[0009] For example, ITK-SNAP is an open source software specifically for medical image segmentation. It uses ITK (Insight Segmentation and Registration Toolkit) as its image processing engine and supports a variety of segmentation algorithms, such as threshold segmentation, region growing, active contours, etc. Users can perform interactive segmentation through a graphical interface and perform post-processing and analysis as needed. Existing problems: For some cases, users are required to perform interactive segmentation, but how to balance automation and user interaction remains a challenge; in terms of performance and efficiency, for large data sets, the performance and efficiency of the segmentation algorithm may be limited.

[0010] Therefore, the technical problems to be solved by the present invention are: to perform relatively objective segmentation of the lesion site based on a new neural network to ensure the segmentation efficiency of large data sets; to improve the image quality of relatively low-quality CT images in mobile operating rooms based on multi-dimensional neural networks to improve the accuracy of image processing results; to establish a neural network prediction model based on a patient database to predict the subsequent development of the disease and output a predicted cerebral hemorrhage image; to include commonly used image processing functions, simplify the operation method, and shorten the user's learning time.

[0011] Defect 2: The supporting systems of some CT-specific equipment are very restrictive and expensive, and are not suitable for current clinical conditions in my country.

[0012] Specific limitations include: the high cost of R&D, manufacturing and maintenance of CT-specific equipment, which leads to its high price; the operation and maintenance of CT equipment requires professional technicians, including radiologists, radiographers and maintenance engineers, which may be affected by the shortage of technicians, thus affecting the normal use and maintenance of the equipment; the operation and interpretation of CT equipment requires relevant personnel to undergo professional training to ensure correct use and diagnosis, and training costs and resources may become limiting factors; some regions may have medical policies and insurance restrictions, resulting in restrictions on the use of some high-end medical equipment, or restrictions on the reimbursement of equipment. Other image processing systems with more complete functions are basically foreign commercial software, and how to perform relevant image processing for lesions has not yet been fully realized in China. Summary of the invention

[0013] In order to overcome the above technical problems, the purpose of the present invention is to provide an intelligent image processing device and method for CT images of acute cerebral hemorrhage. According to the characteristics of CT images of acute cerebral hemorrhage in mobile operating rooms, image processing is performed to adapt to the application of mobile CT, and the supporting system has less restrictions and lower costs.

[0014] The present invention provides the following technical solutions:

[0015] First aspect

[0016] The present invention provides an intelligent image processing method for acute cerebral hemorrhage CT images, which comprises the following steps:

[0017] S1: Database storage, obtaining the original CT image data of acute cerebral hemorrhage in the mobile operating room and storing it in the database;

[0018] S2: Data resolution enhancement: data resolution enhancement is performed on the acquired original CT image data based on a neural network to obtain a CT image with enhanced resolution;

[0019] S3: neural network image segmentation, used to perform neural network-based image segmentation on the enhanced resolution CT image to obtain a segmented image associated with bleeding site information and related pathological information;

[0020] S4: reconstruction, used to reconstruct the segmented image, and reconstruct the patient's head CT data and the segmented lesion location data;

[0021] S5: Establishing a visualization and interactive interface for performing data format conversion and image rendering processing on the three-dimensional model and multi-plane image reconstructed by the reconstruction module, obtaining a head CT image and a three-dimensional model of the bleeding site and a head CT two-dimensional image of different viewing angles for user interactive operation;

[0022] S6: Neural network prediction is used to obtain prediction results based on patient information, pathological information of segmented lesion sites, and information changes in different time periods through improved CNN network training. The prediction results include the development trend of the patient's subsequent condition and the predicted image of the bleeding site.

[0023] In the above implementation, the establishment of the database in S1 is conducive to the preparation of a data set for patients with acute cerebral hemorrhage, and improves the predictive performance of the neural network in the subsequent steps; the data resolution enhancement in S2 can solve the problem of insufficient resolution of mobile CT, greatly improve the scope of application of mobile CT, and provide a feasible solution for future preoperative planning, intraoperative navigation and diagnosis; the establishment of the prediction network in S6 can assist doctors in diagnosis, reduce doctors' time costs, and can also be used for training medical students.

[0024] According to some embodiments, the method of step S2 data resolution enhancement comprises the following steps:

[0025] S2-1: Data preprocessing, de-caging, sorting, renaming and attribute standardization of the original CT image data to obtain preprocessed CT image data;

[0026] S2-2: Multi-dimensional resolution enhancement of CT images, based on neural network, inter-layer expansion and intra-layer resolution improvement of pre-processed CT images.

[0027] According to some embodiments, the neural network in step S3 is selected from a deep convolutional neural network using an attention mechanism.

[0028] According to some embodiments, the neural network training data in step S3 is divided into three training sets, the first training set includes the original bleeding data and normal data; the second training set performs data enhancement on the bleeding data, but the training set only includes the bleeding data and does not include the original data; the third training set performs data enhancement on the bleeding data to keep the data volume of the bleeding data and the normal data balanced, and includes the amplified bleeding data and the original normal data.

[0029] According to some embodiments, the reconstruction in step S4 includes three-dimensional reconstruction and multi-planar reconstruction, wherein the three-dimensional reconstruction is used to complete the spatial position positioning of the bleeding site, and the multi-planar reconstruction is used to obtain two-dimensional images of multiple planes by cutting or projecting on planes at different angles and directions in the three-dimensional medical image data.

[0030] Second aspect

[0031] The present invention also provides an intelligent image processing system for acute cerebral hemorrhage CT images, which comprises:

[0032] A database storage module is used to obtain the original CT image data of acute cerebral hemorrhage in the mobile operating room;

[0033] A data resolution enhancement module performs data resolution enhancement on the acquired original CT image data based on a neural network to obtain a CT image with enhanced resolution;

[0034] A neural network image segmentation module is used to perform neural network-based image segmentation on the enhanced resolution CT image to obtain a segmented image associated with bleeding site information and related pathological information;

[0035] A reconstruction module is used to reconstruct the segmented image, and to reconstruct the patient's head CT data and the segmented lesion location data;

[0036] The visualization and interactive interface module is used to convert the data format and perform image rendering processing on the three-dimensional model and multi-plane image reconstructed by the reconstruction module, so as to obtain the head CT image and the three-dimensional model of the bleeding site and the head CT two-dimensional image of different viewing angles for user interactive operation;

[0037] The neural network prediction module is used to obtain prediction results based on patient information, pathological information of segmented lesion sites, and information changes in different time periods through improved CNN network training. The prediction results include the development trend of the patient's subsequent condition and the predicted image of the bleeding site.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] Aiming at the characteristics of CT images of acute cerebral hemorrhage in mobile operating rooms, the present invention performs image preprocessing, segmentation, reconstruction and prediction in steps to adapt to the application of mobile CT. In the multi-dimensional high-resolution reconstruction part of CT images in mobile operating rooms, the initial image is reconstructed with multi-dimensional high resolution through the residual network model and the generative adversarial network model respectively to improve the accuracy of the subsequent image processing results; in the neural network segmentation image part, the trained model has been verified by a large amount of credible data, which guarantees the accuracy of the system results to the greatest extent. In the multi-plane reconstruction and three-dimensional reconstruction parts, the lesion area can be spatially located in two-dimensional and three-dimensional space in combination with the neural network module, and there is an accurate and intuitive positioning, which can assist in the diagnosis of the disease of mobile CT acute cerebral hemorrhage images. Combined with the personal database neural network intracranial hematoma expansion prediction, the processed data can be stored, and the bleeding site can be subsequently predicted in combination with the neural network to provide an intuitive cerebral hemorrhage prediction image.

[0040] The present invention integrates most commonly used image processing functions, has strong robustness, and proposes a multi-resolution reconstruction network and a disease prediction network, which improves the accuracy and intuitiveness of function implementation and better assists physicians in diagnosis. The present invention has a low learning cost, and medical students can easily use it, thereby improving the level of teaching and training. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a structural block diagram of an intelligent image processing system for acute cerebral hemorrhage CT images provided in Embodiment 2 of the present invention.

[0042] Figure 2 It is an operation flow chart of the intelligent image processing method for acute cerebral hemorrhage CT images provided in Example 1 of the present invention.

[0043] Figure 3 This is an effect diagram of the CT image data provided by the first embodiment of the present invention after other preprocessing but without bed removal processing.

[0044] Figure 4 This is an effect diagram of the CT image data provided in the first embodiment of the present invention after preprocessing, including bed removal processing.

[0045] Figure 5 This is a rendering of a multi-planar reconstruction and a three-dimensional reconstruction without bed removal provided by the first embodiment of the present invention.

[0046] Figure 6 This is a rendering of the three-dimensional reconstruction in step S4 provided in the first embodiment of the present invention.

[0047] Figure 7 This is the visualization and interactive interface provided by the first embodiment of the present invention.

[0048] Figure 8 It is a combination of multiple images segmented in step S3 provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0049] The present invention is described in detail below in conjunction with the embodiments and drawings, but it should be understood that the embodiments and drawings are only used to exemplify the present invention and do not constitute any limitation on the protection scope of the present invention. All reasonable changes and combinations within the scope of the inventive concept of the present invention fall within the protection scope of the present invention.

[0050] CT (computed tomography) usually refers to computer tomography, which uses the different absorption rates of X-rays emitted by different tissues of the human body, and then uses highly sensitive instruments to measure the human body. The measured data is then input into a computer, and the data is reconstructed using an algorithm to obtain a CT image. CT images can display cross-sectional images of the examined part of the human body, thereby performing non-invasive diagnosis of the lesion. CT images are usually exported as DICOM format files. DICOM (Digital Imaging and Communications in Medicine) refers to medical digital imaging and communications, and is an international standard for medical images and related information (ISO 12052). It defines a medical image format that can be used for data exchange with quality that meets clinical needs. Because DICOM images have a higher depth than general images, they can usually only be displayed on professional medical displays that meet the grayscale standard display function to meet clinical diagnostic standards. Compared with ordinary CT, mobile CT has the characteristics of fast scanning speed, fewer scanning layers, thicker scanning layers, lower dose, and lower resolution. In view of the above characteristics, the present invention performs standardized preprocessing on mobile CT images, improves the image resolution through corresponding interpolation algorithms, and can display them on the screen with high precision to meet clinical diagnosis standards. The bleeding area is further segmented to accurately obtain more hidden features such as texture features of the bleeding site and the amount of bleeding. The neural network model extracts feature information from the CT images and can quickly and accurately obtain various information on acute cerebral hemorrhage, providing doctors with evaluation and prediction.

[0051] The present invention will be further described below in conjunction with the accompanying drawings.

[0052] Embodiment 1

[0053] like Figure 2 The present invention proposes an intelligent image processing method for acute cerebral hemorrhage CT images, which comprises the following steps:

[0054] Step S1: Acquire CT image data of acute cerebral hemorrhage in a mobile operating room and store it in a database.

[0055] Step S2: Due to the low resolution of mobile CT scanning, the original CT image data is enhanced in data resolution.

[0056] Specifically, the steps of data resolution enhancement include data preprocessing and multi-dimensional resolution enhancement of CT images.

[0057] In data preprocessing, PYHTON data processing software can be used to perform batch processing operations on a large amount of data, including renaming, sorting, format attribute standardization, and data de-coding by masking. The original CT image data is de-coded, sorted, renamed, and attribute standardized to obtain the pre-processed CT image data. The CT images in png format before and after de-coding are shown in Figure 2. Figure 3 and Figure 4 shown.

[0058] Multi-dimensional resolution enhancement of CT images is performed based on neural networks to expand the inter-layer and improve the intra-layer resolution of pre-processed CT images.

[0059] Multi-dimensional resolution enhancement of CT images is a residual network model that combines the statistical laws between layers of moving CT images and prior knowledge to quickly reconstruct high-resolution images of inter-layer images. For example, each patient undergoes two CT scans with the same number of layers but misaligned in the Z direction, one set as input and the other set as labels. Texture features of moving CT images are extracted through the gray-level co-occurrence matrix and used as an input path. Due to the similarity between adjacent layers of CT images, every 3 layers of images are used as a group of input. The network weights are replaced with a pre-trained VGG network for segmenting the head area, and prior knowledge of the location of the brain is added. Each path passes through a specific convolutional layer and upsampling, connects the output of each path, and finally passes through a fully connected layer to output the generated image of the CT intermediate layer. For the intra-layer resolution, a generative adversarial network model is used to retain the edge information of the lesion and combine the redundant information of other CT scan layers to quickly reconstruct the high-resolution image of the layer. In view of the fact that the low-density and high-density shadows of cerebral hemorrhage are mixed and it is difficult to distinguish the boundaries, the edge-oriented Super Resolution (SR) algorithm is used, that is, the prior knowledge of CT images is added, and the redundant information of the CT scan layer of the adjacent layer and the gradient contour prior algorithm are combined to restore the edge of the high-frequency details. First, the data is preprocessed by bicubic upsampling to reduce unnecessary network operations. The generative adversarial network is divided into two network models, the generator and the discriminator, and the above-mentioned edge-oriented algorithm is placed in the encoding layer of the generator. The feature information of the adjacent CT scan layer extracted by the pre-trained VGG network is input into the network as redundant information. Upsampling is only placed in the last layer to avoid excessive computational costs. The loss function of the generator consists of minimizing the mean square error loss function, the visual perception loss function combined with the grayscale standard display function, and the accuracy of the discriminator to avoid over-smoothing of the reconstructed image. The CT image is expanded between layers and the intra-layer resolution is improved by the method of multi-dimensional resolution enhancement of CT images.

[0060] Step S3: After the neural network image segmentation module obtains the CT image with enhanced resolution, the CT image of the bleeding part needs to be segmented in order to further obtain and analyze the pathological information of the bleeding part. The neural network can segment the cerebral hemorrhage part quickly and accurately.

[0061] Compared with traditional medical image segmentation methods, deep convolutional neural networks can learn more complex and deep features of data and realize end-to-end image segmentation processes. The fully convolutional neural network is modified on the basis of the convolutional neural network, and the image is classified at the pixel level. In order to improve the segmentation effect, the attention mechanism is applied to the convolutional neural network. The attention mechanism is used to obtain the target area that needs to be focused on. The model used in the neural network segmentation module in this embodiment is Attention Unet. Considering the multi-scale range of cerebral hemorrhage and the multi-interference information of intracranial CT images, the attention mechanism is applied to the UNet segmentation network, and the attention module is added to the feature fusion part, which can achieve the focus on the significant area and the suppression of irrelevant background areas. This model is a model with guaranteed accuracy that has been trained with a large amount of data. It can directly segment the preprocessed data without long-term training. The data preprocessing selection uses the PYHTON program to perform batch processing operations on a large amount of data, including renaming, sorting, format attribute standardization, and data de-bedding operations through masks. Specifically, the training of the neural network model includes the following steps:

[0062] First, data enhancement is performed on the data. The training data is divided into bleeding data and normal data, and three training sets are created to train the model. Training set 1 contains the original bleeding data and normal data; training set 2 performs data enhancement on the bleeding data, but the training set only contains bleeding data and does not contain the original data; training set 3 performs data enhancement on the bleeding data to keep the data volume of bleeding data and normal data balanced, and contains the amplified bleeding data and the original normal data.

[0063] Attention Unet neural network model By comparing with other models, Attention Unet has relatively good performance in the segmentation task of acute cerebral hemorrhage images of mobile CT. The specific algorithm flow is as follows: The output of the downsampling convolution layer and the input of the upsampling layer before splicing are respectively fed into a 1×1 convolution to convert them to the same number of channels without changing the size. After the upsampling operation is completed (same size), they are accumulated and passed through ReLU. Through 1x1 convolution and a sigmoId function, the attention score is obtained and assigned to each part of the feature map. The attention map is multiplied by the encoder input to obtain the final output of the attention block. Applying the attention module before feature fusion allows the network to put more weight on features related to jump connections.

[0064] In this way, an accurate segmentation image of the cerebral hemorrhage site is obtained, such as Figure 8 , This figure is a comparison of the label and segmentation result of the brain hemorrhage site. The segmentation label is a reference for network iteration, that is, the standard answer; the segmentation result is the result given by the network after learning.

[0065] Step S4: After performing neural network image segmentation on the preprocessed CT image, multi-plane reconstruction and three-dimensional reconstruction are performed on the segmentation result image according to the preprocessed CT image to complete the spatial location and diagnosis of the cerebral hemorrhage site. Multi-plane refers to the acquisition of two-dimensional images of multiple planes by cutting or projecting on different planes in three-dimensional medical image data. This method allows the three-dimensional structure to be observed and analyzed from different angles and directions, thereby providing more information and details. Generally speaking, it refers to the planes of three perspectives: cross-section, coronal and sagittal.

[0066] Specifically, according to the different storage formats of CT image data, image data in DICOM format and PNG format can be read and imported respectively, and then the data can be converted into geometric metadata. The view direction can be adjusted by setting the attributes of the relevant classes, and multi-plane reconstruction of the cross-section, sagittal and coronal planes of the CT image can be completed respectively, and two-dimensional spatial positioning of different views of the bleeding site can be performed.

[0067] More specifically, after the CT image data is read and imported, the data type is converted for storage, and then the data is extracted for isosurface. According to the MC algorithm, the isosurface is regenerated, the edge curve of the data model is extracted, and finally the image is rendered to generate the head CT scan image and the three-dimensional model of the bleeding site, complete the three-dimensional spatial positioning of the cerebral hemorrhage site, and improve the diagnostic accuracy. Figure 5 Shown is the visualization interface of multi-planar reconstruction and three-dimensional reconstruction, wherein the upper left, upper right, lower left and lower right images are the coronal plane, transverse section, sagittal plane and three-dimensional reconstruction images of multi-planar reconstruction respectively. Figure 6 Shown are the three-dimensional reconstruction effects before and after the bed is removed, with the left picture being the picture after the bed is removed and the right picture being the picture before the bed is removed.

[0068] Step S5: Establish a visualization and interactive interface, perform data visualization, display the reconstruction results on the established visualization and interactive interface, and perform interaction. The existing calculation functions in PYTHON can be called to calculate the hematoma parameters and spatial position coordinates of the segmented parts, obtain quantitative analysis of the image information, and display it on the visualization and interactive interface.

[0069] The above processing results and image data are imported into a personal database and archived in chronological order, and a personal account is established to store the corresponding personal data information.

[0070] like Figure 7 , various functions including neural network calls are performed through buttons on the visualization and interactive interface, and visualization functions are implemented in the right window, and calculation results and predicted text information are displayed in the left window;

[0071] Step S6: Neural network prediction. On the basis of establishing a personal database, the pathological information obtained by the segmentation module is input, and the development of the individual's cerebral hemorrhage site can be predicted through the neural network prediction module. The data with follow-up CT images can be used as training data for the convolutional neural network, i.e., the CNN network. The follow-up CT image data can be used by experienced physicians to outline the final infarct volume as a label. Due to the variability of each CT acquisition, the data is resampled to the same coordinate system, and the two CT acquisition images are aligned. The network input consists of the following three parts: the original CT image data and the subsequent follow-up CT image data; the downsampled CT image data and the subsequent follow-up CT image data; the cerebral hemorrhage segmentation image data; and the metadata (consisting of the acquisition time between the original CT image data and the follow-up CT data, and the interval time between the original image data acquisition and the occurrence of stroke). Each input is used as a path for convolution and upsampling operations, and each path is calculated in parallel. Finally, all path inputs are connected together, and the predicted output image is obtained through three 1×1 convolutional layers. Since the intra-plane resolution and inter-plane resolution of the acquired CT images are different, convolution layers with different numbers of filters are used for each path, so that the output space dimension of each path is consistent. By predicting multiple adjacent voxels at the same time, the computational efficiency and the accuracy of the results are improved.

[0072] Through neural network prediction, the development of intracranial hematoma expansion and cerebral hemorrhage site can be predicted more accurately.

[0073] Embodiment 2

[0074] As an application of the first embodiment, Figure 1 The present embodiment provides an intelligent image processing system for acute cerebral hemorrhage CT images, which includes a database module, a data resolution enhancement module, a neural network image segmentation module, a reconstruction module, a visualization and interface module, and a neural network prediction module.

[0075] The database module is used to import and store the data information of the CT images to be processed and the confidentiality operation of the patient's personal information, including viewing and editing;

[0076] The data resolution enhancement module is used to preprocess the data and enhance the resolution of the CT image to be processed to obtain a CT image with enhanced resolution. It performs statistical analysis and rapid reconstruction through a model based on a residual network and a generative adversarial network.

[0077] The data resolution enhancement module includes a data preprocessing module and a CT image multi-dimensional resolution enhancement module.

[0078] The neural network image segmentation module is used to perform image segmentation on enhanced resolution CT images. The purpose of image segmentation is to reconstruct the bleeding site, obtain relevant pathological information, and calculate the bleeding location, bleeding morphology, and bleeding volume. During segmentation, it is necessary to focus on significant areas and suppress irrelevant background areas.

[0079] The reconstruction module includes a three-dimensional reconstruction module and a multi-planar reconstruction module. The three-dimensional reconstruction module is used to reconstruct the patient's head CT data and the segmented lesion site data. The multi-planar reconstruction module is used to perform multi-planar reconstruction on the patient's head CT data and the segmented lesion site data to obtain planar images of a total of three perspectives, including the coronal plane, sagittal plane and the original cross-section.

[0080] The visualization and interface module is used to convert the data format and perform image rendering processing on the three-dimensional model and multi-plane image reconstructed by the reconstruction module, so as to obtain the head CT image and the three-dimensional model of the bleeding site and the head CT two-dimensional images from three different viewing angles.

[0081] The neural network prediction module obtains the prediction results based on the patient information, the pathological information of the segmented lesion site, and the information changes in different time periods through improved CNN network training. The prediction results include the development trend of the patient's subsequent condition and the predicted image of the bleeding site.

[0082] The above embodiments are only preferred implementations of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent image processing method for acute cerebral hemorrhage CT images, comprising the following steps: S1: Database storage, obtaining the original CT image data of acute cerebral hemorrhage in the mobile operating room and storing it in the database; S2: Data resolution enhancement: data resolution enhancement is performed on the acquired original CT image data based on a neural network to obtain a CT image with enhanced resolution; S3: neural network image segmentation, used to perform neural network-based image segmentation on the enhanced resolution CT image to obtain a segmented image associated with bleeding site information and related pathological information; S4: reconstruction, for performing multi-planar reconstruction and three-dimensional reconstruction on the enhanced resolution CT image and the segmented image to obtain a reconstructed multi-planar image and a three-dimensional model, and complete spatial location and identification of the cerebral hemorrhage site; S5: Establishing a visualization and interactive interface, for performing data format conversion and image rendering processing on the three-dimensional model and multi-plane image reconstructed in step S4, obtaining a three-dimensional head CT image and two-dimensional head CT images of different viewing angles, and establishing a visualization and interactive interface for display; S6: Neural network prediction, which is used to predict the development of an individual's cerebral hemorrhage site based on the patient's personal information, pathological information and the segmented image obtained in step S3, through the established prediction model. The prediction results include the patient's subsequent disease development trend and the predicted image of the bleeding site; the prediction model is trained using a CNN network.

2. The intelligent image processing method for acute cerebral hemorrhage CT images according to claim 1, characterized in that: Step S2: The method for enhancing data resolution comprises the following steps: S2-1: Data preprocessing, de-caging, sorting, renaming and attribute standardization of the original CT image data to obtain preprocessed CT image data; S2-2: Multi-dimensional resolution enhancement of CT images, based on neural network, inter-layer expansion and intra-layer resolution improvement of pre-processed CT images.

3. The intelligent image processing method for acute cerebral hemorrhage CT images according to claim 1, characterized in that: The neural network in step S3 is selected from a deep convolutional neural network using an attention mechanism.

4. The intelligent image processing method for acute cerebral hemorrhage CT images according to claim 1, characterized in that: The neural network training data in step S3 is divided into three training sets. The first training set includes the original bleeding data and normal data; the second training set performs data enhancement on the bleeding data, but the training set only includes the bleeding data and does not include the original data; the third training set performs data enhancement on the bleeding data to keep the data volume of the bleeding data and the normal data balanced, and includes the amplified bleeding data and the original normal data.

5. The intelligent image processing method for acute cerebral hemorrhage CT images according to claim 1, characterized in that: The reconstruction in step S4 includes three-dimensional reconstruction and multi-planar reconstruction. The three-dimensional reconstruction is used to complete the spatial position positioning of the bleeding site, and the multi-planar reconstruction is used to obtain two-dimensional images of multiple planes by cutting or projecting the enhanced resolution CT image and the segmented image on planes of different angles and directions.

6. An intelligent image processing system for acute cerebral hemorrhage CT images, characterized in that: It includes: A database storage module is used to obtain the original CT image data of acute cerebral hemorrhage in the mobile operating room; A data resolution enhancement module performs data resolution enhancement on the acquired original CT image data based on a neural network to obtain a CT image with enhanced resolution; A neural network image segmentation module is used to perform neural network-based image segmentation on the enhanced resolution CT image to obtain a segmented image associated with bleeding site information and related pathological information; A reconstruction module, used for performing multi-planar reconstruction and three-dimensional reconstruction on the enhanced resolution CT image and the segmented image, obtaining a reconstructed multi-planar image and a three-dimensional model, and completing the spatial location and identification of the cerebral hemorrhage site; A visualization and interactive interface module is used to perform data format conversion and image rendering processing on the three-dimensional model and multi-plane image reconstructed by the reconstruction module, obtain a three-dimensional head CT image and two-dimensional head CT images of different viewing angles, and establish a visualization and interactive interface for display; The neural network prediction module is used to predict the development of an individual's cerebral hemorrhage site based on the patient's personal information, pathological information and the segmented image obtained in step S3, and through the established prediction model, the prediction result includes the patient's subsequent development trend of the disease and the predicted image of the bleeding site; the prediction model is trained using a CNN network.

Images