CT image intelligent identification and analysis system based on deep learning

Abstract

The invention relates to the technical field of image analysis, in particular to a CT image intelligent recognition and analysis system based on deep learning, and the system comprises an uploading module which is used for uploading a brain CT image of a diagnosed patient and marking the brain CT image of the patient; the importing module is used for importing the patient brain CT image which is uploaded and marked in the uploading module into a ResNet model for training; according to the method, the prior patient brain CT image is imported into the deep learning model for training, so that intelligent diagnosis is provided for the patient through the deep learning model and the patient brain CT image, the reception pressure of medical staff is effectively relieved, and the diagnosis efficiency is improved when the patient is diagnosed through the deep learning module. According to the method, the name of the disease of the patient can be accurately output, matching is carried out based on prior data, a recommended treatment scheme is provided for the patient, and the convenience of diagnosis and treatment of brain diseases at the present stage is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and particularly relates to a CT image intelligent recognition and analysis system based on deep learning. Background Art

[0002] A brain CT image is an image generated after a tomographic scan of the brain using X-rays and processed by a computer. It can clearly show the structures of brain tissues, bones, etc., helping doctors detect brain lesions such as tumors, hemorrhages, infarcts, etc., and plays a crucial role in the diagnosis of brain diseases, the evaluation of the condition, and the formulation of treatment plans.

[0003] The invention patent application with the application number 202410286917.6 proposes a method for recognizing head CT images based on deep learning, including: for the original head CT image and the rough head CT image obtained by reducing the original head CT image, respectively performing an extraction operation on the lesion elements based on the preliminary recognition unit in the pre-trained voxel analysis model to obtain the corresponding original lesion structure and rough lesion structure; the lesion element is the brain lesion area included in the original head image; integrating the original lesion structure and the rough lesion structure to obtain an integrated lesion structure; the integrated lesion structure is used to represent the key structure data of the brain lesion area; for the original head CT image and the integrated lesion structure, performing an extraction operation on the lesion elements based on the fine recognition unit according to the recurrent neural network in the voxel analysis model to obtain a lesion structure extraction result corresponding to the lesion element.

[0004] This application points out the problem that "traditional head CT image analysis mainly relies on the experience of radiologists for manual interpretation, which not only takes a long time but is also easily affected by individual experience differences, resulting in certain subjectivity and uncertainty in the diagnosis results. In recent years, deep learning technology has made remarkable progress in the fields of image recognition, natural language processing, etc. Especially in the field of medical image analysis, deep learning-based methods can automatically learn complex features in images, improving the accuracy and efficiency of disease diagnosis. Nevertheless, due to the high-dimensional characteristics of head CT images and the complexity of brain lesion areas, it poses challenges to the accurate recognition and localization of brain lesions".

[0005] However, existing brain CT image processing technologies often focus on the accuracy and clarity of images, all of which are for medical staff to better observe brain CT images, while the technology that can read brain CT images and diagnose the patients from whom the brain CT images are sourced is still immature.

[0006] Therefore, a CT image intelligent recognition and analysis system based on deep learning is proposed. Summary of the Invention

[0007] In view of the above-mentioned drawbacks of the prior art, the present invention provides an intelligent recognition and analysis system for CT images based on deep learning, which solves the technical problems proposed in the above-mentioned background art.

[0008] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0009] An intelligent recognition and analysis system for CT images based on deep learning, comprising:

[0010] An upload module for uploading the brain CT images of diagnosed patients and annotating the brain CT images of patients; an import module for importing the brain CT images of patients uploaded and completed annotation in the upload module into the ResNet model for training; a collection module for collecting the current brain CT images of patients captured by the CT image collection device and sending the brain CT images of patients to the trained ResNet model; a receiving module for receiving the results of the processing of the brain CT images of patients sent by the collection module to the ResNet model in the ResNet model; an analysis module for analyzing the matching degree between the brain CT images of patients collected by the collection module and the brain CT images of each patient imported into the ResNet model for training; a recommendation module for traversing the analysis results of the brain CT matching degree of patients in the analysis module and generating a treatment plan recommendation for the patients from whom the brain CT images collected by the collection module are sourced based on the analysis results.

[0011] Furthermore, the annotation content of the brain CT images of patients in the upload module includes the name of the diagnosed disease of the patient and the image of the lesion area in the brain CT image of the patient;

[0012] Sub-modules are provided under the upload module and the collection module, including:

[0013] A cropping unit for receiving the brain CT images of patients, setting the target size of image cropping, and cropping the brain CT images of patients according to the set target size of image cropping;

[0014] A normalization unit for receiving the brain CT images of patients cropped by the cropping unit and performing normalization processing on the brain CT images of patients so that all brain CT images of patients are within a preset same gray value range;

[0015] Among them, the cropping unit receives one brain CT image of a patient each time it runs, forwards it to the normalization unit synchronously after cropping the received brain CT image of the patient, refreshes and runs synchronously, receives the brain CT image of the patient again and performs the cropping and forwarding operations, and the normalization unit runs synchronously following the cropping unit.

[0016] Furthermore, the target size of image cropping in the cropping unit is user-defined by the system user. When performing the cropping operation on the patient's brain CT image, it follows:

[0017] Convert the patient's brain CT image into a grayscale image, perform adaptive threshold segmentation using the Otsu algorithm, and convert the image into a binary image;

[0018] On the binary image, use the findContours function in OpenCV to find all the contours in the image;

[0019] Among all the found contours, select the contour with the largest area as the contour of the brain region;

[0020] For the determined contour of the brain region, use the boundingRect function to calculate the bounding box of this contour;

[0021] According to the calculated bounding box parameters, crop the part containing the brain region from the original image to obtain the cropped image;

[0022] Among them, after the patient's brain CT image is collected, a scaling operation is synchronously performed according to the preset image size, so that the sizes of all the completed cropped patient's brain CT images output from the cropping unit are equal.

[0023] Furthermore, the normalization processing logic for the patient's brain CT image in the normalization unit is expressed as:

[0024] Perform truncation processing on the pixel values of the patient's brain CT image:

[0025]

[0026] The output result of the pixel in the patient's brain CT image:

[0027]

[0028] In the formula: X trunc is the pixel value after truncation processing; X is the value of the pixel in the patient's brain CT image; is the quantile; X norm is the pixel value after pixel normalization processing; new max 、new min are the minimum and maximum values that the pixel values of the normalized patient's brain CT image can reach, that is, the preset grayscale value interval;

[0029] Among them, X trunc In the calculation stage, all the pixels in the patient's brain CT image are sorted based on their pixel values, and the default values of p1 and p2 are 0.01 and 0.99, That is, it represents the value of the pixel at the 1% position in the pixel queue sorted based on pixel values. Similarly.

[0030] Furthermore, during the running stage of the import module, the patient's brain CT images that have completed cropping and normalization processing are divided into a training set, a validation set, and a test set.

[0031] The ResNet model is trained using the training set. Through parameter adjustment of the ResNet model, the model completes the recognition and classification of different lesions in the patient's brain CT images included in the training set. During the training process, the validation set is used to evaluate the performance of the model to adjust the hyperparameters. After the training is completed, the test set is input into the trained model to obtain the accuracy, precision, recall rate, F1 value, and mean squared error of the model. When the accuracy, precision, recall rate, F1 value, and mean squared error of the model all meet the preset qualified judgment interval, any one of the SGD, Adagrad, and Adadelta algorithms is further used to minimize the loss function of the model.

[0032] Among them, the parameter adjustment of the ResNet model includes the weights of the convolutional kernels, the sizes of the convolutional kernels, and the biases.

[0033] Furthermore, the results of processing the patient's brain CT images received by the receiving module in the ResNet model include: disease name, lesion area image, lesion area size, and brain tissue density of the lesion area.

[0034] Among them, when the results of processing the patient's brain CT images received by the receiving module in the ResNet model are incomplete, the patient's diagnosis result is an asymptomatic patient.

[0035] Furthermore, the matching degree between the patient's brain CT images in the analysis module is determined based on the similarity of the patient's brain CT images.

[0036]

[0037] In the formula: SIMM is the similarity of the patient's brain CT images; α is the weight; SIMM avg is the average structural similarity of two patient's brain CT images; is the distance between the corresponding GLCM feature vectors of two patient's brain CT images;

[0038] Among them, α ∈ [0, 1], SIMM avg When calculating, the two patient's brain CT images for similarity calculation are segmented with the same operation. After calculating the structural similarity of each pair of corresponding sub-images obtained by segmentation, the sum of all the structural similarity calculation results is averaged, that is, SIMM avg .

[0039] Further, a sub-module is provided under the analysis module, including:

[0040] A queue unit, configured to obtain the matching degree between each patient's brain CT image imported into the ResNet model for training in the analysis module and the patient's brain CT image collected by the collection module, and sort the brain CT images of each patient imported into the ResNet model for training in descending order based on the matching degree to generate a patient brain CT image queue.

[0041] Further, the treatment plan recommended for the patient in the recommendation module is: the treatment plan used by the patient from whom the brain CT image at the forefront of the patient brain CT image queue is sourced.

[0042] Further, the upload module is connected to the import module and the collection module through wireless network interaction. The upload module is connected to a cropping unit and a normalization unit through wireless network interaction under the collection module. The collection module is connected to a receiving module, an analysis module, and a recommendation module through wireless network interaction. The analysis module is connected to a queue unit through wireless network interaction at a lower level.

[0043] Adopting the technical solution provided by the present invention, compared with the known public technology, it has the following beneficial effects:

[0044] The present invention provides a CT image intelligent recognition and analysis system based on deep learning. During the operation stage of the system, prior patient brain CT images are imported into the deep learning model for training, so as to provide intelligent diagnosis for patients by combining the deep learning model with the patient brain CT images, effectively alleviating the reception pressure of medical staff. When diagnosing patients through the deep learning module, it can output the name of the disease suffered by the patient relatively accurately, and match based on prior data, and at the same time provide a recommended treatment plan for the patient, effectively improving the convenience of the diagnosis and treatment of brain diseases at the present stage. Brief Description of the Drawings

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0046] Figure 1 It is a schematic structural diagram of a CT image intelligent recognition and analysis system based on deep learning. Detailed Embodiments

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0048] The following further describes the present invention with reference to embodiments.

[0049] Embodiment:

[0050] An intelligent CT image recognition and analysis system based on deep learning in this embodiment, as Figure 1 shown, includes:

[0051] An upload module 1, configured to upload the brain CT images of diagnosed patients and label the brain CT images of patients.

[0052] An import module 2, configured to import the brain CT images of patients uploaded and labeled in the upload module 1 into the ResNet model for training.

[0053] During the operation stage of the import module 2, the brain CT images of patients that have completed cropping and normalization processing are divided into a training set, a validation set, and a test set.

[0054] Use the training set to train the ResNet model. Through parameter adjustment of the ResNet model, the model completes the recognition and classification of different lesions in the brain CT images of patients included in the training set. During the training process, the validation set is used to evaluate the performance of the model to adjust the hyperparameters. After the training is completed, the test set is input into the trained model to obtain the accuracy, precision, recall rate, F1 value, and mean square error of the model. When the accuracy, precision, recall rate, F1 value, and mean square error of the model all meet the preset qualified determination interval, any one of the SGD, Adagrad, and Adadelta algorithms is further used to minimize the loss function of the model.

[0055] Among them, the parameter adjustment of the ResNet model includes the weights of the convolutional kernels, the sizes of the convolutional kernels, and the biases.

[0056] An acquisition module 3, configured to acquire the current brain CT images of patients captured by the CT image acquisition device and send the brain CT images of patients to the trained ResNet model.

[0057] The annotation content of the brain CT images of patients in the upload module 1 includes the names of the diagnosed diseases of the patients and the images of the lesion areas in the brain CT images of the patients.

[0058] The upload module 1 and the acquisition module 3 have sub - modules at the lower level, including:

[0059] The cropping unit 11 is used to receive the patient's brain CT image, set the target size for image cropping, and crop the patient's brain CT image according to the set target size for image cropping;

[0060] The normalization unit 12 is used to receive the patient's brain CT image cropped by the cropping unit 11, and perform normalization processing on the patient's brain CT image so that all patient's brain CT images are within a preset same gray - value range;

[0061] Among them, the cropping unit 11 receives one patient's brain CT image each time it runs. After cropping the received patient's brain CT image, it synchronously forwards it to the normalization unit 12. The cropping unit 11 synchronously refreshes its operation, receives the patient's brain CT image again and performs the cropping and forwarding operations, and the normalization unit 12 runs synchronously following the cropping unit 11;

[0062] The target size for image cropping in the cropping unit 11 is user - defined by the system - side user. When performing the cropping operation on the patient's brain CT image, it follows:

[0063] Convert the patient's brain CT image into a grayscale image, perform adaptive threshold segmentation using the Otsu algorithm, and convert the image into a binary image;

[0064] On the binary image, use the findContours function in OpenCV to find all the contours in the image;

[0065] Among all the found contours, select the contour with the largest area as the contour of the brain region;

[0066] For the determined contour of the brain region, use the boundingRect function to calculate the bounding box of the contour;

[0067] According to the calculated bounding - box parameters, crop the part containing the brain region from the original image to obtain the cropped image;

[0068] Among them, after the patient's brain CT image is acquired, it synchronously performs a scaling operation according to the preset image size so that the sizes of all the cropped patient's brain CT images output from the cropping unit 11 are equal;

[0069] The logic for normalizing the patient's brain CT image in the normalization unit 12 is expressed as:

[0070] Perform truncation processing on the pixel values of the patient's brain CT image:

[0071]

[0072] Output results of pixels in the patient's brain CT image:

[0073]

[0074] Where: X trunc is the pixel value after truncation processing; X is the value of the pixel in the patient's brain CT image; is the quantile; X norm is the pixel value after pixel normalization processing; new max and new min are the minimum and maximum values that the pixel values of the patient's brain CT image can reach after normalization, that is, the preset gray value interval;

[0075] Among them, X trunc In the calculation stage, all pixels in the patient's brain CT image are sorted based on their pixel values. The default values of p1 and p2 are 0.01 and 0.99. That is, it represents the value of the pixel at the 1% position in the pixel queue sorted based on the pixel value. Similarly;

[0076] Through the above formula, the normalization processing operation of the patient's brain CT image provides specific operation logic limitations.

[0077] The receiving module 4 is used to receive the results of the processing of the patient's brain CT image sent by the acquisition module 3 to the ResNet model in the ResNet model;

[0078] The analysis module 5 is used to analyze the matching degree between the patient's brain CT image collected by the acquisition module 3 and each patient's brain CT image imported into the ResNet model for training;

[0079] The matching degree between the patient's brain CT images in the analysis module 5 is determined based on the similarity of the patient's brain CT images;

[0080]

[0081] Where: SIMM is the similarity of the patient's brain CT image; α is the weight; SIMM avg is the average structural similarity of two patient's brain CT images; is the distance between the corresponding GLCM feature vectors of two patient's brain CT images;

[0082] Among them, α ∈ [0, 1], SIMM avg In the calculation, the two patient's brain CT images for similarity calculation are segmented by the same operation. After calculating the structural similarity of each pair of corresponding sub-images obtained by segmentation, the sum of all the structural similarity calculation results is averaged, that is, SIMMavg ;

[0083] Through the above logical formula calculation, a specified digital calculation logic is provided for the matching degree between the brain CT images of patients, providing further operation support for the operation of the recommendation module 6 in this embodiment of the system.

[0084] The analysis module 5 is provided with sub-modules at a lower level, including:

[0085] The queue unit 51 is used to obtain the matching degree between each brain CT image of the patient imported into the ResNet model for training in the analysis module 5 and the brain CT image of the patient collected by the collection module 3, and sort the brain CT images of each patient imported into the ResNet model for training in descending order based on the matching degree to generate a queue of brain CT images of patients;

[0086] The recommendation module 6 is used to traverse the analysis results of the brain CT matching degree of the patients in the analysis module 5, and generate a treatment plan recommendation for the patient from whom the brain CT image collected by the collection module 3 is sourced based on the analysis results;

[0087] The treatment plan recommendation generated for the patient in the recommendation module 6 is: the treatment plan used by the patient from whom the brain CT image at the forefront of the patient brain CT image queue is sourced;

[0088] The upload module 1 is connected to the import module 2 and the collection module 3 through wireless network interaction. The upload module 1 is connected to the cropping unit 11 and the normalization unit 12 through wireless network interaction at a lower level of the collection module 3. The collection module is connected to the receiving module 4, the analysis module 5 and the recommendation module 6 through wireless network interaction. The analysis module 5 is connected to the queue unit 51 through wireless network interaction at a lower level.

[0089] In this embodiment, the upload module 1 runs to upload the brain CT images of the diagnosed patients, annotate the brain CT images of the patients, and the import module 2 runs later to import the brain CT images of the patients uploaded and annotated in the upload module 1 into the ResNet model for training. The acquisition module 3 further acquires the brain CT images of the patients currently captured by the CT image acquisition device, and sends the brain CT images of the patients to the trained ResNet model. The cropping unit 11 receives the brain CT images of the patients in real time during the operation of the upload module 1 and the acquisition module 3, sets the target size for image cropping, and crops the brain CT images of the patients according to the set target size for image cropping. The normalization unit 12 synchronously receives the brain CT images of the patients cropped by the cropping unit 11, and performs normalization processing on the brain CT images of the patients, so that all the brain CT images of the patients are in a preset same gray value interval. The receiving module 4 further receives the results of the processing of the brain CT images of the patients sent by the acquisition module 3 to the ResNet model in the ResNet model, and then the analysis module 5 analyzes the matching degree between the brain CT images of the patients acquired by the acquisition module 3 and the brain CT images of each patient imported into the ResNet model for training. The queue unit 51 synchronously obtains the matching degree between the brain CT images of each patient imported into the ResNet model for training in the analysis module 5 and the brain CT images of the patients acquired by the acquisition module 3, and arranges the brain CT images of each patient imported into the ResNet model for training in descending order based on the matching degree to generate a queue of brain CT images of the patients. Finally, the recommendation module 6 traverses the analysis results of the brain CT image matching degree in the analysis module 5, and generates a treatment plan recommendation for the patients from whom the brain CT images acquired by the acquisition module 3 are sourced based on the analysis results.

[0090] Through the operation of the above system, combined with the deep learning module, the recognition and diagnosis of brain CT images are provided for the patients, and a treatment plan recommendation is provided for the patients, effectively improving the convenience of the patients' medical treatment. At the same time, the burden on medical staff for receiving patients is reduced, and the effect of intelligent medical services is further realized.

[0091] As Figure 1 shown, the results of the processing of the brain CT images of the patients received by the receiving module 4 running in the ResNet model include: disease name, lesion area image, lesion area size, and brain tissue density of the lesion area;

[0092] Among them, when the results of the processing of the brain CT images of the patients received by the receiving module 4 running are incomplete, the patient diagnosis result is an asymptomatic patient.

[0093] Through the above settings, the content output by the ResNet model after processing the brain CT images of the patients is further defined.

[0094] In the above embodiments, during the operation of the system, the prior patient brain CT images are imported into the deep learning model for training, so as to provide intelligent diagnosis for the patient by combining the patient brain CT images through the deep learning model, effectively alleviating the pressure on medical staff during patient reception. When diagnosing the patient through the deep learning module, it can accurately output the name of the disease the patient suffers from, and match based on prior data, while providing a recommended treatment plan for the patient, effectively improving the convenience of diagnosing and treating brain diseases at the present stage.

[0095] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A CT image intelligent recognition and analysis system based on deep learning, characterized in that: include: An uploading module (1) is used to upload the brain CT images of the diagnosed patients and to annotate the brain CT images of the patients; An import module (2) is used to import the patient's brain CT image uploaded and annotated in the upload module (1) into the ResNet model for training; An acquisition module (3) is used to acquire a CT image of the patient's brain currently captured by a CT image acquisition device, and send the CT image of the patient's brain to the trained ResNet model; A receiving module (4) is used for receiving the result of processing the patient's brain CT image in the ResNet model sent by the acquisition module (3) to the ResNet model; An analysis module (5) is used to analyze the matching degree between the patient's brain CT image acquired by the acquisition module (3) and each patient's brain CT image imported into the ResNet model for training; The recommendation module (6) is used to traverse the patient brain CT matching analysis results in the analysis module (5), and generate a treatment plan recommendation for the patient whose brain CT image is collected in the collection module (3) based on the analysis results.

2. The CT image intelligent recognition and analysis system based on deep learning according to claim 1, characterized in that: The annotation content of the patient's brain CT image in the uploading module (1) includes the name of the patient's confirmed disease and the image of the lesion area in the patient's brain CT image; The upload module (1) and the acquisition module (3) are provided with submodules at the lower level, including: A cropping unit (11) is used to receive a patient's brain CT image, set an image cropping target size, and crop the patient's brain CT image according to the set image cropping target size; A normalization unit (12) is used to receive the patient's brain CT images cropped by the cropping unit (11), and perform normalization processing on the patient's brain CT images so that all the patient's brain CT images are in the same preset grayscale value range; The cropping unit (11) receives a patient's brain CT image each time it runs, and after cropping the received patient's brain CT image, it synchronously forwards it to the normalization unit (12). The cropping unit (11) synchronously refreshes and runs, receives the patient's brain CT image again and performs cropping and forwarding operations, and the normalization unit (12) runs synchronously with the cropping unit (11).

3. The CT image intelligent recognition and analysis system based on deep learning according to claim 2, characterized in that: The image cropping target size in the cropping unit (11) is customized by the system end user. When the cropping operation is performed on the patient's brain CT image, it complies with: The patient's brain CT image was converted into a grayscale image, and the Otsu algorithm was used for adaptive threshold segmentation to convert the image into a binary image; On a binary image, use the findContours function in OpenCV to find all contours in the image; Among all the contours found, the contour with the largest area is selected as the contour of the brain region; For the determined brain region contour, use the boundingRect function to calculate the bounding box of the contour; According to the calculated bounding box parameters, a portion including the brain region is cropped from the original image to obtain a cropped image; After the acquisition of the patient's brain CT image is completed, a scaling operation is synchronously performed according to a preset image size, so that the sizes of all the cropped patient's brain CT images output from the cropping unit (11) are equal.

4. The CT image intelligent recognition and analysis system based on deep learning according to claim 2, characterized in that: The normalization processing logic of the patient's brain CT image in the normalization unit (12) is expressed as: The pixel values ​​of the patient's brain CT image are truncated: Output of pixels in a CT image of a patient's brain: Where: X trunc is the pixel value after truncation; X is the pixel value in the patient's brain CT image; is the quantile; X norm is the pixel value after pixel normalization; new max 、new min It is the minimum and maximum values ​​that the pixel value of the patient's brain CT image can reach after normalization, that is, the preset gray value interval; Among them, X trunc In the calculation stage, all pixels in the patient's brain CT image are sorted based on their pixel values. The default values ​​of p1 and p2 are 0.01 and 0.

99. That is, it represents the value of the pixel at the 1% position in the pixel queue sorted based on pixel value. Same reason.

5. The CT image intelligent recognition and analysis system based on deep learning according to claim 1, characterized in that: During the operation phase of the import module (2), the cropped and normalized patient brain CT images are divided into a training set, a validation set, and a test set; The ResNet model is trained using the training set. By adjusting the parameters of the ResNet model, the model can complete the recognition and classification of different lesions in the patient's brain CT images included in the training set. During the training process, the validation set is used to evaluate the performance of the model to adjust the hyperparameters. After the training is completed, the test set is input into the trained model to obtain the accuracy, precision, recall, F1 value, and mean square error of the model. When the accuracy, precision, recall, F1 value, and mean square error of the model all meet the preset qualified judgment interval, any one of the algorithms of SGD, Adagrad, and Adadelta is further used to minimize the loss function of the model. Among them, the parameter adjustment of the ResNet model includes convolution kernel weight, convolution kernel size, and bias.

6. The CT image intelligent recognition and analysis system based on deep learning according to claim 1, characterized in that: The receiving module (4) processes the received patient's brain CT image in the ResNet model, and the result includes: disease name, lesion area image, lesion area size, and brain tissue density in the lesion area; When the result of processing the received brain CT image of the patient in the ResNet model is incomplete, the patient is diagnosed as an asymptomatic patient.

7. The CT image intelligent recognition and analysis system based on deep learning according to claim 1, characterized in that: The matching degree between the patient's brain CT images in the analysis module (5) is determined based on the similarity of the patient's brain CT images; Where: SIMM is the similarity of the patient's brain CT images; α is the weight; SIMM avg is the average structural similarity of the two patients’ brain CT images; is the distance between the corresponding GLCM feature vectors of the two patients’ brain CT images; Among them, α∈[0,1], SIMM avg During the calculation, the same segmentation operation is performed on the two brain CT images of the patient for similarity calculation. After calculating the structural similarity of the corresponding sub-images obtained by each segmentation, all the structural similarity calculation results are summed and averaged, that is, SIMM avg .

8. The CT image intelligent recognition and analysis system based on deep learning according to claim 1, characterized in that: The analysis module (5) is provided with submodules at the lower level, including: The queue unit (51) is used to obtain the matching degree between each patient's brain CT image imported into the ResNet model for training in the analysis module (5) and the patient's brain CT image acquired by the acquisition module (3), and to arrange each patient's brain CT image imported into the ResNet model for training in descending order based on the matching degree, so as to generate a patient's brain CT image queue.

9. The CT image intelligent recognition and analysis system based on deep learning according to claim 1, characterized in that: The treatment plan recommendation generated for the patient in the recommendation module (6) is: the treatment plan used by the patient whose brain CT image is the source of the patient at the front of the patient's brain CT image queue.

10. The CT image intelligent recognition and analysis system based on deep learning according to claim 1, characterized in that: The upload module (1) is interactively connected to the import module (2) and the acquisition module (3) via a wireless network; the upload module (1) is interactively connected to the cropping unit (11) and the normalization unit (12) at the lower level of the acquisition module (3) via a wireless network; the acquisition module is interactively connected to the receiving module (4), the analysis module (5) and the recommendation module (6) via a wireless network; and the analysis module (5) is interactively connected to the lower level of the queue unit (51) via a wireless network.

Images