Neurovascular tumor recognition method and system based on image recognition

Abstract

The invention relates to the field of tumor recognition, in particular to a neurovascular tumor recognition method and system based on image recognition. A neurovascular tumor recognition system based on image recognition comprises a sample acquisition module, a patient data acquisition module and a neurovascular tumor recognition module. According to the method, the correlation between different MRI image modalities and tumor types, malignancy and invasiveness is calculated, global statistical features and deep learning features are fused, the most representative and distinguishable features required by all recognition tasks can be recognized, and the fusion strategy not only improves the tumor recognition ability of the model, but also improves the recognition efficiency of the model. And the generalization ability of the model to tumors of different types and characteristics is enhanced, so that the model is more flexible and effective in actual clinical application.

Description

Technical Field

[0001] The present invention relates to the field of tumor recognition, and in particular to a method and system for neurovascular tumor recognition based on image recognition. Background Art

[0002] In modern medical diagnosis, early identification and accurate diagnosis of neurovascular tumors are crucial to the treatment and prognosis of patients. With the development of medical imaging technology, magnetic resonance imaging (MRI) has become an important means of detecting and diagnosing neurovascular tumors. MRI can provide rich anatomical and functional information, but its interpretation requires professional radiologists and neurologists, is time-consuming, and has a certain degree of subjectivity.

[0003] Traditional neurovascular tumor identification methods mainly rely on the doctor's experience and visual judgment, which has certain limitations. First, the doctor's diagnostic ability is limited by his personal experience and knowledge level, and there may be diagnostic differences between different doctors. Second, manual interpretation of MRI images is time-consuming, which is not conducive to rapid diagnosis and emergency treatment. In addition, even experienced doctors may make misdiagnoses for complex or rare cases.

[0004] Therefore, developing a neurovascular tumor recognition method and system that integrates multimodal MRI image information and fuses global statistical features and deep learning features has important research value and clinical significance for improving the accuracy and robustness of recognition. Summary of the invention

[0005] The present invention calculates the correlation between different MRI image modalities and tumor type, malignancy and invasiveness, integrates global statistical features and deep learning features, and can identify the most representative and discriminative features required for each recognition task. This fusion strategy not only improves the model's ability to recognize tumors, but also enhances the model's generalization ability for tumors of different types and characteristics, making it more flexible and effective in actual clinical applications.

[0006] A neurovascular tumor recognition method based on image recognition, comprising: Acquire several patient data samples, each of which includes a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set, and a fluid-attenuated inversion recovery image set obtained by a historical patient using magnetic resonance imaging, and use the tumor detection result, tumor type recognition result, tumor malignancy recognition result, and tumor invasiveness recognition result of the historical patient as labels of the patient data sample; All acquired patient data samples are used to calculate representative correlations between T1-weighted images, T2-weighted images, T1-weighted enhanced images, and fluid-attenuated inversion recovery images and tumor type, tumor malignancy, and tumor invasiveness; a neurovascular tumor recognition model is constructed based on the calculated representative correlations, and the neurovascular tumor recognition model is trained using patient data samples; For the patient to be identified, magnetic resonance imaging is used to obtain the patient's T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set; the acquired T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of the patient to be identified are used as input of the neurovascular tumor recognition model, and the tumor detection result, tumor type recognition result, tumor malignancy recognition result and tumor invasiveness recognition result of the patient to be identified are output.

[0007] Preferably, the tumor detection result is 1 or 0, 1 indicates that the patient has a neurovascular tumor, and 0 indicates that it does not exist; the tumor type identification result is a positive integer in the range of 1 to Z, Z represents the number of all neurovascular tumor types, and each neurovascular tumor type corresponds to a value; the tumor malignancy identification result is 1 or 0, 1 indicates that the patient's neurovascular tumor is a malignant tumor, and 0 indicates that it is a benign tumor; the tumor invasiveness identification result is 1 or 0, 1 indicates that the patient's neurovascular tumor is invasive, and 0 indicates that it is non-invasive; wherein, if the tumor detection result is 0, the tumor type identification result, the tumor malignancy identification result, and the tumor invasiveness identification result are all invalid.

[0008] Preferably, representative correlations between T1-weighted images, T2-weighted images, T1-weighted enhanced images, and fluid-attenuated inversion recovery images and tumor types, tumor malignancy, and tumor invasiveness are calculated using all acquired patient data samples, and the specific operations are as follows: Step 1: Filter out all patient data samples Patient data samples with tumor detection results of 1, among the screened Select any one patient data sample from the patient data samples; Step 2: For any T1-weighted image in the T1-weighted image set of the current patient data sample, use the image processing algorithm and the gray-level co-occurrence matrix to obtain the statistical indicator set of the T1-weighted image, which includes the pixel mean, pixel standard deviation, pixel skewness, pixel kurtosis, contrast, homogeneity, energy and entropy of the T1-weighted image; based on the statistical indicator set of all T1-weighted images in the acquired T1-weighted image set, calculate the average value of each statistical indicator to obtain the comprehensive statistical indicator set of the T1-weighted image set, which includes the average pixel mean, average pixel standard deviation, average pixel skewness, average pixel kurtosis, average contrast, average homogeneity, average energy and average entropy of the T1-weighted image set; Step 3: for any one of the T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of the current patient data sample, perform the same operation as in step 2 to obtain the comprehensive statistical index set of the T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set in the current patient data sample; Step 4: Traverse all the patient data samples except those selected in step 1 -1 patient data sample, repeat the same operations from step 2 to step 3, and finally obtain Comprehensive statistical index set of T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of patient data samples; Step 5: Based on the obtained The comprehensive statistical index set of a T1-weighted image set is calculated using the formula: ; Calculate the correlation coefficient , =1, 2, ..., 8, which represent the pixel mean, pixel standard deviation, pixel skewness, pixel kurtosis, contrast, homogeneity, energy and entropy in statistical indicators respectively; =1, 2, 3, respectively representing the tumor type, tumor malignancy, and tumor invasiveness in the tumor recognition task; =1, 2, …, ; Indicates The first T1-weighted image in the patient data sample Comprehensive statistical indicators; express T1-weighted image set The average value of comprehensive statistical indicators; Indicates In the patient data sample The recognition results of the tumor recognition task; express In the patient data sample The average value of the recognition results of the tumor recognition task; the correlation coefficient That is, the first The comprehensive statistical indicators and The correlation coefficient of all T1-weighted image sets obtained from the calculation is The maximum value is selected as the representative correlation between T1-weighted images and tumor types; the correlation coefficient of all T1-weighted image sets obtained is calculated. The maximum value is selected as the representative correlation between T1-weighted images and tumor malignancy; the correlation coefficient of all T1-weighted image sets obtained by calculation is The maximum value was selected as the representative correlation between T1-weighted images and tumor invasiveness; Step 6: Target The same operation as step five is performed on any one of the T2-weighted image sets, T1-weighted enhanced image sets and fluid-attenuated inversion recovery image sets of the patient data samples, and finally representative correlations between the T2-weighted images, T1-weighted enhanced images and fluid-attenuated inversion recovery images and tumor type, tumor malignancy and tumor invasiveness are obtained respectively.

[0009] Preferably, the neurovascular tumor recognition model is established based on a CNN model, including an input layer, a feature extraction layer, a tumor detection layer, and a task branching layer; The input layer includes four input channels, and the four input channels are used to receive a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set, and a fluid-attenuated inversion recovery image set, respectively; The feature extraction layer is used to output the first feature map, the second feature map, the third feature map and the fourth feature map corresponding to the T1-weighted image set, the T2-weighted image set, the T1-weighted enhanced image set and the fluid-attenuated inversion recovery image set respectively; The tumor detection layer includes a feature flattening layer, a first fully connected layer, a first output layer and a mask mechanism layer. The feature flattening layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and obtain a flattened feature vector through a flattening operation; the first fully connected layer is used to further comprehensively process the flattened feature vector to form a tumor detection feature vector; the first output layer is used to output the tumor detection result; the mask mechanism layer is used to control the execution of subsequent tasks according to the tumor detection result. If the tumor detection result is 1, the task branch layer is allowed to continue to execute; if the tumor detection result is 0, the task branch layer is no longer executed; The task branch layer is used to output tumor type recognition results, tumor malignancy recognition results, and tumor invasiveness recognition results respectively.

[0010] Preferably, the feature extraction layer is connected to four input channels, channel one is connected to the first convolution layer and the first pooling layer, the first convolution layer is used to extract the features of the T1-weighted image, and perform dimensionality reduction through the first pooling layer to obtain a first feature map; channel two is connected to the second convolution layer and the second pooling layer, the second convolution layer is used to extract the features of the T2-weighted image, and perform dimensionality reduction through the second pooling layer to obtain a second feature map; channel three is connected to the third convolution layer and the third pooling layer, the third convolution layer is used to extract the features of the T1-weighted enhanced image, and perform dimensionality reduction through the third pooling layer to obtain a third feature map; channel four is connected to the fourth convolution layer and the fourth pooling layer, the fourth convolution layer is used to extract the features of the fluid-attenuated inversion recovery image, and perform dimensionality reduction through the fourth pooling layer to obtain a fourth feature map.

[0011] Preferably, the task branch layer includes branch one, branch two and branch three; branch one includes a first fusion layer, a second fully connected layer and a second output layer; the first fusion layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and apply the representative correlation between the tumor type and the T1-weighted image, the T2-weighted image, the T1-weighted enhanced image and the fluid-attenuated inversion recovery image to perform weighted fusion on the first feature map, the second feature map, the third feature map and the fourth feature map to obtain the first fusion feature; the second fully connected layer is used to receive the first fusion feature and further extract features; the second output layer is used to output the tumor type recognition result; Branch two includes a second fusion layer, a third fully connected layer and a third output layer; the second fusion layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and apply the representative correlation of tumor malignancy with T1-weighted image, T2-weighted image, T1-weighted enhanced image and fluid attenuation inversion recovery image to perform weighted fusion on the first feature map, the second feature map, the third feature map and the fourth feature map to obtain the second fusion feature; the third fully connected layer is used to receive the second fusion feature and further extract features; the third output layer is used to output the tumor malignancy recognition result; Branch three includes a third fusion layer, a fourth fully connected layer and a fourth output layer; the third fusion layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and apply the representative correlation of tumor invasiveness with T1-weighted images, T2-weighted images, T1-weighted enhanced images and fluid-attenuated inversion recovery images to perform weighted fusion on the first feature map, the second feature map, the third feature map and the fourth feature map to obtain the third fusion feature; the fourth fully connected layer is used to receive the third fusion feature and further extract features; the fourth output layer is used to apply and output the tumor invasiveness recognition result.

[0012] Preferably, the neurovascular tumor recognition model is trained using patient data samples, and the specific operations are as follows: All acquired patient data samples are divided into a tumor recognition training set and a tumor recognition verification set. The neurovascular tumor recognition model is trained using the tumor recognition training set, and then the neurovascular tumor recognition model is verified using the tumor recognition verification set to obtain the verification results. The training conditions are set to determine whether the verification results meet the training conditions. If so, the trained neurovascular tumor recognition model is output; if not, the neurovascular tumor recognition model is continued to be trained using the tumor recognition training set.

[0013] A neurovascular tumor recognition system based on image recognition, comprising: A sample acquisition module is used to acquire a number of patient data samples, each of which includes a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set, and a fluid-attenuated inversion recovery image set obtained by a historical patient using magnetic resonance imaging, and the tumor detection result, tumor type recognition result, tumor malignancy recognition result, and tumor invasiveness recognition result of the historical patient are used as labels for the patient data sample; A patient data acquisition module, used for acquiring a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set and a fluid-attenuated inversion recovery image set of a patient to be identified by using magnetic resonance imaging; The neurovascular tumor recognition module includes a correlation calculation unit, a model construction unit and a neurovascular tumor recognition unit; the correlation calculation unit is used to use all acquired patient data samples to respectively calculate representative correlations between T1-weighted images, T2-weighted images, T1-weighted enhanced images and fluid-attenuated inversion recovery images and tumor types, tumor malignancy and tumor invasiveness; the model construction unit is used to construct a neurovascular tumor recognition model and train the neurovascular tumor recognition model through patient data samples; the neurovascular tumor recognition unit is used to use the acquired T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of the patient to be recognized as inputs of the neurovascular tumor recognition model, and output tumor detection results, tumor type recognition results, tumor malignancy recognition results and tumor invasiveness recognition results of the patient to be recognized.

[0014] The present invention has the following advantages: The present invention calculates the correlation between different MRI image modalities and tumor type, malignancy and invasiveness, integrates global statistical features and deep learning features, and can identify the most representative and discriminative features required for each recognition task. This fusion strategy not only improves the model's ability to recognize tumors, but also enhances the model's generalization ability for tumors of different types and characteristics, making it more flexible and effective in actual clinical applications.

[0015] The present invention can capture the characteristics of tumors from different angles by comprehensively utilizing four MRI image modalities: T1-weighted imaging, T2-weighted imaging, T1-weighted enhanced imaging, and fluid-attenuated inversion recovery imaging, thereby providing more comprehensive tumor information. Combined with convolutional neural networks, the model can automatically learn and extract complex features in images, which not only reduces dependence on physician experience, but also greatly improves the objectivity and automation level of the recognition process, making the diagnosis more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the structure of a neurovascular tumor recognition system based on image recognition adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0018] Embodiment 1, a method for identifying neurovascular tumors based on image recognition, comprising: Acquire several patient data samples, each of which contains a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set, and a fluid-attenuated inversion recovery image set obtained by a historical patient using magnetic resonance imaging, wherein the T1-weighted image set is an image showing the anatomical structure, the T2-weighted image set is an image sensitive to water content, the T1-weighted enhanced image set is a T1-weighted image after contrast agent injection, which is used to observe blood vessels and lesions, and the fluid-attenuated inversion recovery image set is an image that suppresses cerebrospinal fluid signals and enhances the lesion area; the use of multimodal data allows the system to combine the advantages of different imaging technologies to provide a more comprehensive tumor feature analysis; the tumor detection results, tumor type recognition results, tumor malignancy recognition results, and tumor invasiveness recognition results of the historical patient are used as labels for the patient data samples; All acquired patient data samples are used to calculate representative correlations between T1-weighted images, T2-weighted images, T1-weighted enhanced images, and fluid-attenuated inversion recovery images and tumor type, tumor malignancy, and tumor invasiveness; a neurovascular tumor recognition model is constructed based on the calculated representative correlations, and the neurovascular tumor recognition model is trained using patient data samples; The benefit of calculating the correlation between each recognition task and T1-weighted images, T2-weighted images, T1-weighted enhanced images, and fluid-attenuated inversion recovery images is that it can reveal the contribution and importance of different MRI imaging techniques in tumor recognition, help identify which imaging parameters are most sensitive to the characterization of tumor characteristics, and thus provide guidance for building a more accurate neurovascular tumor recognition model; this analysis helps to optimize feature selection, enhance the predictive ability of the model, and improve the accuracy of diagnosis; For the patient to be identified, magnetic resonance imaging is used to obtain the patient's T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set; the acquired T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of the patient to be identified are used as input of the neurovascular tumor recognition model, and the tumor detection result, tumor type recognition result, tumor malignancy recognition result and tumor invasiveness recognition result of the patient to be identified are output.

[0019] The tumor detection result is 1 or 0, 1 indicates that the patient has a neurovascular tumor, and 0 indicates that it does not exist; the tumor type recognition result is a positive integer in the range of 1 to Z, Z represents the number of all neurovascular tumor types, and each neurovascular tumor type corresponds to a value; the tumor malignancy recognition result is 1 or 0, 1 indicates that the patient's neurovascular tumor is a malignant tumor, and 0 indicates that it is a benign tumor; the tumor invasiveness recognition result is 1 or 0, 1 indicates that the patient's neurovascular tumor is invasive, and 0 indicates that it is non-invasive; among them, if the tumor detection result is 0, the tumor type recognition result, tumor malignancy recognition result and tumor invasiveness recognition result are all invalid.

[0020] All acquired patient data samples were used to calculate representative correlations between T1-weighted images, T2-weighted images, T1-weighted enhanced images, and fluid-attenuated inversion recovery images and tumor type, tumor malignancy, and tumor invasiveness. The specific operations are as follows: Step 1: Filter out all patient data samples Patient data samples with tumor detection results of 1, among the screened Select any one patient data sample from the patient data samples; Step 2: For any T1-weighted image in the T1-weighted image set of the current patient data sample, use the image processing algorithm and the gray-level co-occurrence matrix to obtain the statistical indicator set of the T1-weighted image, which includes the pixel mean, pixel standard deviation, pixel skewness, pixel kurtosis, contrast, homogeneity, energy and entropy of the T1-weighted image; based on the statistical indicator set of all T1-weighted images in the acquired T1-weighted image set, calculate the average value of each statistical indicator to obtain the comprehensive statistical indicator set of the T1-weighted image set, which includes the average pixel mean, average pixel standard deviation, average pixel skewness, average pixel kurtosis, average contrast, average homogeneity, average energy and average entropy of the T1-weighted image set; Step 3: for any one of the T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of the current patient data sample, perform the same operation as in step 2 to obtain the comprehensive statistical index set of the T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set in the current patient data sample; Step 4: Traverse all the patient data samples except those selected in step 1 -1 patient data sample, repeat the same operations from step 2 to step 3, and finally obtain Comprehensive statistical index set of T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of patient data samples; Step 5: Based on the obtained The comprehensive statistical index set of a T1-weighted image set is calculated using the formula: ; Calculate the correlation coefficient , =1, 2, ..., 8, which represent the pixel mean, pixel standard deviation, pixel skewness, pixel kurtosis, contrast, homogeneity, energy and entropy in statistical indicators respectively; =1, 2, 3, respectively representing the tumor type, tumor malignancy, and tumor invasiveness in the tumor recognition task; =1, 2, …, ; Indicates The first T1-weighted image in the patient data sample Comprehensive statistical indicators; express T1-weighted image set The average value of comprehensive statistical indicators; Indicates In the patient data sample The recognition results of the tumor recognition task; express In the patient data sample The average value of the recognition results of the tumor recognition task; the correlation coefficient That is, the first The comprehensive statistical indicators and The correlation coefficient of all T1-weighted image sets obtained from the calculation is The maximum value is selected as the representative correlation between T1-weighted images and tumor types; the correlation coefficient of all T1-weighted image sets obtained is calculated. The maximum value is selected as the representative correlation between T1-weighted images and tumor malignancy; the correlation coefficient of all T1-weighted image sets obtained by calculation is The maximum value was selected as the representative correlation between T1-weighted images and tumor invasiveness; Step 6: Target The same operation as step five is performed on any one of the T2-weighted image sets, T1-weighted enhanced image sets and fluid-attenuated inversion recovery image sets of the patient data samples, and finally representative correlations between the T2-weighted images, T1-weighted enhanced images and fluid-attenuated inversion recovery images and tumor type, tumor malignancy and tumor invasiveness are obtained respectively.

[0021] The neurovascular tumor recognition model is built based on the CNN model, including input layer, feature extraction layer, tumor detection layer and task branching layer; The input layer includes four input channels, and the four input channels are used to receive a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set, and a fluid-attenuated inversion recovery image set, respectively; The feature extraction layer is used to output the first feature map, the second feature map, the third feature map and the fourth feature map corresponding to the T1-weighted image set, the T2-weighted image set, the T1-weighted enhanced image set and the fluid-attenuated inversion recovery image set respectively; The tumor detection layer includes a feature flattening layer, a first fully connected layer, a first output layer and a mask mechanism layer. The feature flattening layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and obtain a flattened feature vector through a flattening operation; the first fully connected layer is used to further comprehensively process the flattened feature vector to form a tumor detection feature vector; the first output layer is used to output the tumor detection result; the mask mechanism layer is used to control the execution of subsequent tasks according to the tumor detection result. If the tumor detection result is 1, the task branch layer is allowed to continue to execute; if the tumor detection result is 0, the task branch layer is no longer executed; The tumor detection task is executed first, and its output results are used as the conditions for controlling the execution of subsequent tasks; this avoids unnecessary calculations on tumor-free samples, improves computational efficiency, and reduces the risk of misclassification; The task branch layer is used to output tumor type recognition results, tumor malignancy recognition results, and tumor invasiveness recognition results respectively.

[0022] The feature extraction layer is connected to four input channels. Channel one is connected to the first convolution layer and the first pooling layer. The first convolution layer is used to extract the features of the T1-weighted image, and performs dimensionality reduction through the first pooling layer to obtain the first feature map; channel two is connected to the second convolution layer and the second pooling layer. The second convolution layer is used to extract the features of the T2-weighted image, and performs dimensionality reduction through the second pooling layer to obtain the second feature map; channel three is connected to the third convolution layer and the third pooling layer. The third convolution layer is used to extract the features of the T1-weighted enhanced image, and performs dimensionality reduction through the third pooling layer to obtain the third feature map; channel four is connected to the fourth convolution layer and the fourth pooling layer. The fourth convolution layer is used to extract the features of the fluid-attenuated inversion recovery image, and performs dimensionality reduction through the fourth pooling layer to obtain the fourth feature map.

[0023] The task branch layer includes branch one, branch two and branch three; branch one includes a first fusion layer, a second fully connected layer and a second output layer; the first fusion layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and apply the representative correlation between the tumor type and the T1-weighted image, the T2-weighted image, the T1-weighted enhanced image and the fluid-attenuated inversion recovery image to perform weighted fusion on the first feature map, the second feature map, the third feature map and the fourth feature map to obtain the first fusion feature; the second fully connected layer is used to receive the first fusion feature and further extract features; the second output layer is used to output the tumor type recognition result; Branch two includes a second fusion layer, a third fully connected layer and a third output layer; the second fusion layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and apply the representative correlation of tumor malignancy with T1-weighted image, T2-weighted image, T1-weighted enhanced image and fluid attenuation inversion recovery image to perform weighted fusion on the first feature map, the second feature map, the third feature map and the fourth feature map to obtain the second fusion feature; the third fully connected layer is used to receive the second fusion feature and further extract features; the third output layer is used to output the tumor malignancy recognition result; Branch three includes a third fusion layer, a fourth fully connected layer and a fourth output layer; the third fusion layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and apply the representative correlation of tumor invasiveness with T1-weighted images, T2-weighted images, T1-weighted enhanced images and fluid-attenuated inversion recovery images to perform weighted fusion on the first feature map, the second feature map, the third feature map and the fourth feature map to obtain the third fusion feature; the fourth fully connected layer is used to receive the third fusion feature and further extract features; the fourth output layer is used to apply and output the tumor invasiveness recognition result; Each task branch (tumor type recognition, malignancy recognition, and invasiveness recognition) is assigned different weights according to its correlation with MRI images of different modalities. This mechanism enables the model to selectively focus on and utilize image information of different modalities according to the different needs of the task, rather than simply averaging all features. In this way, the model can flexibly adjust and optimize the utilization of features according to different task characteristics to improve recognition accuracy. For example, tumor type recognition may be more dependent on the texture information of T2-weighted images, while malignancy recognition may be more dependent on the contrast of T1-weighted enhanced images. Through weighted fusion, the recognition ability of each task can be maximized.

[0024] The neurovascular tumor recognition model is trained using patient data samples. The specific operations are as follows: All acquired patient data samples are divided into a tumor recognition training set and a tumor recognition verification set. The neurovascular tumor recognition model is trained using the tumor recognition training set, and then the neurovascular tumor recognition model is verified using the tumor recognition verification set to obtain the verification results. The training conditions are set to determine whether the verification results meet the training conditions. If so, the trained neurovascular tumor recognition model is output; if not, the neurovascular tumor recognition model is continued to be trained using the tumor recognition training set.

[0025] Embodiment 2, a neurovascular tumor recognition system based on image recognition, such as Figure 1 As shown, including: A sample acquisition module is used to acquire a number of patient data samples, each of which includes a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set, and a fluid-attenuated inversion recovery image set obtained by a historical patient using magnetic resonance imaging, and the tumor detection result, tumor type recognition result, tumor malignancy recognition result, and tumor invasiveness recognition result of the historical patient are used as labels for the patient data sample; A patient data acquisition module, used for acquiring a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set and a fluid-attenuated inversion recovery image set of a patient to be identified by using magnetic resonance imaging; The neurovascular tumor recognition module includes a correlation calculation unit, a model construction unit and a neurovascular tumor recognition unit; the correlation calculation unit is used to use all acquired patient data samples to respectively calculate representative correlations between T1-weighted images, T2-weighted images, T1-weighted enhanced images and fluid-attenuated inversion recovery images and tumor types, tumor malignancy and tumor invasiveness; the model construction unit is used to construct a neurovascular tumor recognition model and train the neurovascular tumor recognition model through patient data samples; the neurovascular tumor recognition unit is used to use the acquired T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of the patient to be recognized as inputs of the neurovascular tumor recognition model, and output tumor detection results, tumor type recognition results, tumor malignancy recognition results and tumor invasiveness recognition results of the patient to be recognized.

[0026] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A neurovascular tumor recognition method based on image recognition, characterized in that: include: Acquire several patient data samples, each of which includes a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set, and a fluid-attenuated inversion recovery image set obtained by a historical patient using magnetic resonance imaging, and use the tumor detection result, tumor type recognition result, tumor malignancy recognition result, and tumor invasiveness recognition result of the historical patient as labels of the patient data sample; All acquired patient data samples are used to calculate representative correlations between T1-weighted images, T2-weighted images, T1-weighted enhanced images, and fluid-attenuated inversion recovery images and tumor type, tumor malignancy, and tumor invasiveness; a neurovascular tumor recognition model is constructed based on the calculated representative correlations, and the neurovascular tumor recognition model is trained using patient data samples; For the patient to be identified, magnetic resonance imaging is used to obtain the patient's T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set; the acquired T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of the patient to be identified are used as input of the neurovascular tumor recognition model, and the tumor detection result, tumor type recognition result, tumor malignancy recognition result and tumor invasiveness recognition result of the patient to be identified are output.

2. The neurovascular tumor recognition method based on image recognition according to claim 1, characterized in that: The tumor detection result is 1 or 0, 1 indicates that the patient has a neurovascular tumor, and 0 indicates that it does not exist; the tumor type recognition result is a positive integer in the range of 1 to Z, Z represents the number of all neurovascular tumor types, and each neurovascular tumor type corresponds to a value; the tumor malignancy recognition result is 1 or 0, 1 indicates that the patient's neurovascular tumor is a malignant tumor, and 0 indicates that it is a benign tumor; the tumor invasiveness recognition result is 1 or 0, 1 indicates that the patient's neurovascular tumor is invasive, and 0 indicates that it is non-invasive; among them, if the tumor detection result is 0, the tumor type recognition result, tumor malignancy recognition result and tumor invasiveness recognition result are all invalid.

3. The neurovascular tumor recognition method based on image recognition according to claim 2, characterized in that: All acquired patient data samples were used to calculate representative correlations between T1-weighted images, T2-weighted images, T1-weighted enhanced images, and fluid-attenuated inversion recovery images and tumor type, tumor malignancy, and tumor invasiveness. The specific operations are as follows: Step 1: Filter out all patient data samples Patient data samples with tumor detection results of 1, among the screened Select any one patient data sample from the patient data samples; Step 2: For any T1-weighted image in the T1-weighted image set of the current patient data sample, use the image processing algorithm and the gray-level co-occurrence matrix to obtain the statistical indicator set of the T1-weighted image, which includes the pixel mean, pixel standard deviation, pixel skewness, pixel kurtosis, contrast, homogeneity, energy and entropy of the T1-weighted image; based on the statistical indicator set of all T1-weighted images in the acquired T1-weighted image set, calculate the average value of each statistical indicator to obtain the comprehensive statistical indicator set of the T1-weighted image set, which includes the average pixel mean, average pixel standard deviation, average pixel skewness, average pixel kurtosis, average contrast, average homogeneity, average energy and average entropy of the T1-weighted image set; Step 3: for any one of the T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of the current patient data sample, perform the same operation as in step 2 to obtain the comprehensive statistical index set of the T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set in the current patient data sample; Step 4: Traverse all the patient data samples except those selected in step 1 -1 patient data sample, repeat the same operations from step 2 to step 3, and finally obtain Comprehensive statistical index set of T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of patient data samples; Step 5: Based on the acquisition The comprehensive statistical index set of a T1-weighted image set is calculated using the formula: ； Calculate the correlation coefficient , =1, 2, ..., 8, which represent the pixel mean, pixel standard deviation, pixel skewness, pixel kurtosis, contrast, homogeneity, energy and entropy in statistical indicators respectively; =1, 2, 3, respectively representing the tumor type, tumor malignancy, and tumor invasiveness in the tumor recognition task; =1, 2, …, ; Indicates The first T1-weighted image in the patient data sample Comprehensive statistical indicators; express T1-weighted image set The average value of comprehensive statistical indicators; Indicates In the patient data sample The recognition results of the tumor recognition task; express In the patient data sample The average value of the recognition results of the tumor recognition task; the correlation coefficient That is, the first The comprehensive statistical indicators and The relevance of the tumor recognition task; The correlation coefficients of all T1-weighted image sets were calculated. The maximum value is selected as the representative correlation between T1-weighted images and tumor types; the correlation coefficient of all T1-weighted image sets obtained is calculated. The maximum value is selected as the representative correlation between T1-weighted images and tumor malignancy; the correlation coefficient of all T1-weighted image sets obtained by calculation is The maximum value was selected as the representative correlation between T1-weighted images and tumor invasiveness; Step 6: Target The same operation as step five is performed on any one of the T2-weighted image sets, T1-weighted enhanced image sets and fluid-attenuated inversion recovery image sets of the patient data samples, and finally representative correlations between the T2-weighted images, T1-weighted enhanced images and fluid-attenuated inversion recovery images and tumor type, tumor malignancy and tumor invasiveness are obtained respectively.

4. The method for identifying neurovascular tumors based on image recognition according to claim 3, characterized in that: The neurovascular tumor recognition model is built based on the CNN model, including input layer, feature extraction layer, tumor detection layer and task branching layer; The input layer includes four input channels, and the four input channels are used to receive a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set, and a fluid-attenuated inversion recovery image set, respectively; The feature extraction layer is used to output the first feature map, the second feature map, the third feature map and the fourth feature map corresponding to the T1-weighted image set, the T2-weighted image set, the T1-weighted enhanced image set and the fluid-attenuated inversion recovery image set respectively; The tumor detection layer includes a feature flattening layer, a first fully connected layer, a first output layer and a mask mechanism layer. The feature flattening layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and obtain a flattened feature vector through a flattening operation; the first fully connected layer is used to further comprehensively process the flattened feature vector to form a tumor detection feature vector; the first output layer is used to output the tumor detection result; the mask mechanism layer is used to control the execution of subsequent tasks according to the tumor detection result. If the tumor detection result is 1, the task branch layer is allowed to continue to execute; if the tumor detection result is 0, the task branch layer is no longer executed; The task branch layer is used to output tumor type recognition results, tumor malignancy recognition results, and tumor invasiveness recognition results respectively.

5. The method for identifying neurovascular tumors based on image recognition according to claim 4, characterized in that: The feature extraction layer is connected to four input channels. Channel one is connected to the first convolution layer and the first pooling layer. The first convolution layer is used to extract the features of the T1-weighted image, and performs dimensionality reduction through the first pooling layer to obtain the first feature map; channel two is connected to the second convolution layer and the second pooling layer. The second convolution layer is used to extract the features of the T2-weighted image, and performs dimensionality reduction through the second pooling layer to obtain the second feature map; channel three is connected to the third convolution layer and the third pooling layer. The third convolution layer is used to extract the features of the T1-weighted enhanced image, and performs dimensionality reduction through the third pooling layer to obtain the third feature map; channel four is connected to the fourth convolution layer and the fourth pooling layer. The fourth convolution layer is used to extract the features of the fluid-attenuated inversion recovery image, and performs dimensionality reduction through the fourth pooling layer to obtain the fourth feature map.

6. The method for identifying neurovascular tumors based on image recognition according to claim 5, characterized in that: The task branch layer includes branch one, branch two and branch three; branch one includes a first fusion layer, a second fully connected layer and a second output layer; the first fusion layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and apply the representative correlation between the tumor type and the T1-weighted image, the T2-weighted image, the T1-weighted enhanced image and the fluid-attenuated inversion recovery image to perform weighted fusion on the first feature map, the second feature map, the third feature map and the fourth feature map to obtain the first fusion feature; The second fully connected layer is used to receive the first fusion feature and further extract features; The second output layer is used to output tumor type recognition results; Branch two includes a second fusion layer, a third fully connected layer and a third output layer; the second fusion layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and apply the representative correlation of tumor malignancy with the T1-weighted image, the T2-weighted image, the T1-weighted enhanced image and the fluid-attenuated inversion recovery image to perform weighted fusion on the first feature map, the second feature map, the third feature map and the fourth feature map to obtain a second fusion feature; The third fully connected layer is used to receive the second fusion feature and further extract features; The third output layer is used to output the tumor malignancy recognition results; Branch three includes a third fusion layer, a fourth fully connected layer and a fourth output layer; the third fusion layer is used to receive the first feature map, the second feature map, the third feature map and the fourth feature map output by the feature extraction layer, and apply the representative correlation of tumor invasiveness with T1-weighted images, T2-weighted images, T1-weighted enhanced images and fluid-attenuated inversion recovery images to perform weighted fusion on the first feature map, the second feature map, the third feature map and the fourth feature map to obtain a third fusion feature; The fourth fully connected layer is used to receive the third fusion feature and further extract features; The fourth output layer is used to apply the output of tumor aggressiveness recognition results.

7. The method for identifying neurovascular tumors based on image recognition according to claim 6, characterized in that: The neurovascular tumor recognition model is trained using patient data samples. The specific operations are as follows: All acquired patient data samples are divided into a tumor recognition training set and a tumor recognition verification set. The neurovascular tumor recognition model is trained using the tumor recognition training set, and then the neurovascular tumor recognition model is verified using the tumor recognition verification set to obtain the verification results. The training conditions are set to determine whether the verification results meet the training conditions. If so, the trained neurovascular tumor recognition model is output; if not, the neurovascular tumor recognition model is continued to be trained using the tumor recognition training set.

8. A neurovascular tumor recognition system based on image recognition, characterized in that: The system is applied to a neurovascular tumor recognition method based on image recognition as described in any one of claims 1 to 7, comprising: A sample acquisition module is used to acquire a number of patient data samples, each of which includes a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set, and a fluid-attenuated inversion recovery image set obtained by a historical patient using magnetic resonance imaging, and the tumor detection result, tumor type recognition result, tumor malignancy recognition result, and tumor invasiveness recognition result of the historical patient are used as labels for the patient data sample; A patient data acquisition module, used for acquiring a T1-weighted image set, a T2-weighted image set, a T1-weighted enhanced image set and a fluid-attenuated inversion recovery image set of a patient to be identified by using magnetic resonance imaging; The neurovascular tumor recognition module includes a correlation calculation unit, a model construction unit and a neurovascular tumor recognition unit; the correlation calculation unit is used to use all acquired patient data samples to respectively calculate representative correlations between T1-weighted images, T2-weighted images, T1-weighted enhanced images and fluid-attenuated inversion recovery images and tumor types, tumor malignancy and tumor invasiveness; the model construction unit is used to construct a neurovascular tumor recognition model and train the neurovascular tumor recognition model through patient data samples; the neurovascular tumor recognition unit is used to use the acquired T1-weighted image set, T2-weighted image set, T1-weighted enhanced image set and fluid-attenuated inversion recovery image set of the patient to be recognized as inputs of the neurovascular tumor recognition model, and output tumor detection results, tumor type recognition results, tumor malignancy recognition results and tumor invasiveness recognition results of the patient to be recognized.

Images