A pituitary tumor dynamic risk assessment and monitoring method and system

Abstract

The present invention discloses a pituitary tumor dynamic risk assessment and monitoring method and system, which relates to the field of medical technology. The method comprises: obtaining pituitary tumor dynamic monitoring data, data processing and feature extraction, establishing a dynamic monitoring prediction model, dynamically monitoring and predicting pituitary tumors based on a pre-trained BiGRU neural network attention model, and calculating a risk index. The present invention realizes a more objective and accurate dynamic risk prediction of pituitary tumors based on a pre-trained BiGRU neural network attention model and combines MRI imaging data of pituitary tumors and peripheral arterial blood vessels and hormone level data. Blood samples are clustered according to hormone level data based on an improved FCM clustering algorithm, and a stratified sampling method is combined to select representative data for feature extraction, thereby improving the accuracy of subsequent predictions.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a method and system for dynamic risk assessment and monitoring of pituitary tumors. Background Art

[0002] Pituitary tumors are a group of tumors that arise from the residual cells of the anterior and posterior pituitary gland and the craniopharyngeal epithelium. About 10% of intracranial tumors have obvious clinical symptoms, with slightly more men than women. Pituitary tumors usually occur in young and middle-aged people, and often affect the patient's growth and development, reproductive function, learning and work ability. Clinical manifestations include abnormal hormone secretion syndrome, tumor compression of surrounding pituitary tissues, pituitary stroke and other manifestations of anterior pituitary dysfunction. The existing medical system's monitoring and treatment of pituitary tumors still relies on imaging-based methods. Existing research institutions are using artificial intelligence to predict pituitary tumors, but they are still in the preliminary stage. Only MRI images are considered in the prediction, and the prediction effect is poor. Summary of the invention

[0003] In order to solve the technical problems of dynamic risk assessment and monitoring of pituitary tumors in the prior art, the present invention provides a method and system for dynamic risk assessment and monitoring of pituitary tumors.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for dynamic risk assessment and monitoring of pituitary tumors, characterized by comprising:

[0006] Obtain dynamic monitoring data of pituitary tumors, including MRI imaging data and hormone level data;

[0007] Data processing and feature extraction, including MRI image data processing and feature extraction and hormone level data processing and feature extraction, to obtain MRI image data features and hormone level data features;

[0008] Establish a dynamic monitoring prediction model to dynamically monitor and predict pituitary tumors based on the pre-trained BiGRU neural network attention model;

[0009] The MRI image data processing and feature extraction includes pituitary tumor image feature extraction and blood flow information feature extraction, the pituitary tumor image features include size, shape, and position, and the blood flow information features include hemodynamic features obtained based on the internal carotid artery and the anterior cerebral artery, including blood flow, blood flow resistance, blood volume, and vascular morphology;

[0010] The hormone level data processing includes clustering the blood samples according to the hormone level data based on the improved FCM clustering algorithm;

[0011] The risk index was calculated and input into the dynamic monitoring prediction model based on the MRI image data characteristics and hormone level data characteristics obtained above to obtain the pituitary tumor risk index.

[0012] Furthermore, the MRI image data includes MRI image data of a pituitary tumor and peripheral arterial blood vessels; the peripheral arterial blood vessels include an internal carotid artery and an anterior cerebral artery.

[0013] Furthermore, the improved FCM clustering algorithm comprises the following steps:

[0014] Calculate the initial membership matrix based on the random membership values;

[0015] The first cluster center is calculated according to the following formula : ;in, N is the sample size, The fuzziness is m Time j Samples Belong to i The degree of class, The fuzziness is m Time k The samples belong to i The degree of class, m For fuzziness.

[0016] The second cluster center is calculated according to the following formula : ;in, , Respectively k samples and t samples, For the k The samples belong to i The degree of class, For the t The samples belong to i The degree of class, C is the number of sample clusters;

[0017] Combined with the second cluster center And the membership value , update the Lagrange multiplier using the following formula : ;in, For the j samples; The fuzziness is m -1 hour j Samples Belong to i the degree of class;

[0018] Combined with the second cluster center and the Lagrange multiplier, the membership matrix is ​​updated according to the following formula: ; Further obtain the updated membership matrix U.

[0019] Repeat the steps from calculating the first cluster center to updating the membership matrix until the change in the membership value in the latest two generations is less than a given threshold or the number of generations reaches a set value, then stop the generation, defuzzify the latest membership matrix, and obtain the clustering result.

[0020] Furthermore, the pre-trained BiGRU neural network attention model includes a pre-training layer, a time domain convolution layer, a bidirectional gated recurrent network layer, and an attention mechanism layer.

[0021] Furthermore, the time domain convolution layer introduces variable dilated convolution.

[0022] Furthermore, the time domain convolution layer is composed of multiple stacked residual blocks, each residual block includes two causal dilated convolution layers, a batch normalization layer and an activation function layer.

[0023] The present invention also provides a pituitary tumor dynamic risk assessment and monitoring system, based on the pituitary tumor dynamic risk assessment and monitoring method as described above, which comprises:

[0024] A pituitary tumor dynamic monitoring data acquisition module is used to acquire pituitary tumor dynamic monitoring data, including MRI image data and hormone level data; the MRI image data includes MRI image data of the pituitary tumor and peripheral arterial blood vessels; the peripheral arterial blood vessels include the internal carotid artery and the anterior cerebral artery.

[0025] The data processing and feature extraction module is used to process and extract features of MRI image data and process hormone level data to obtain MRI image data features and hormone level data features; the MRI image data processing and features include pituitary tumor image features and blood flow information features, the pituitary tumor image features include size, shape, and position, and the blood flow information features include hemodynamic features obtained from the internal carotid artery and the anterior cerebral artery, including blood flow, blood flow resistance, blood volume, and vascular morphology. The hormone level data processing includes clustering blood samples according to hormone level data based on an improved FCM clustering algorithm.

[0026] A dynamic monitoring prediction model building module, which is used to dynamically monitor and predict pituitary tumors based on a pre-trained BiGRU neural network attention model;

[0027] The risk index calculation module is used to obtain the pituitary tumor risk index by inputting the dynamic monitoring prediction model based on the MRI image data characteristics and hormone level data characteristics obtained above.

[0028] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which program instructions of a dynamic risk assessment and monitoring method for pituitary tumors are stored. The program instructions of the dynamic risk assessment and monitoring method for pituitary tumors can be executed by one or more processors to implement the steps of the dynamic risk assessment and monitoring method for pituitary tumors as described above.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] The present invention realizes dynamic risk assessment and monitoring of pituitary tumors. Based on the pre-trained BiGRU neural network attention model and combined with MRI imaging data of pituitary tumors and peripheral arterial blood vessels and hormone level data, a more objective and accurate dynamic risk prediction of pituitary tumors is achieved. Blood samples are clustered according to hormone level data based on the improved FCM clustering algorithm, and the stratified sampling method is combined to select representative data for feature extraction, thereby improving the accuracy of subsequent predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0032] Figure 1 It is a schematic flow chart of a method for dynamic risk assessment and monitoring of pituitary tumors according to an embodiment of the present application;

[0033] Figure 2 is a schematic diagram of a pre-trained BiGRU neural network attention model according to an embodiment of the present application;

[0034] Figure 3 It is a schematic diagram of the structure of the dilated convolution according to an embodiment of the present application. DETAILED DESCRIPTION

[0035] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0036] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0037] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention. The illustrations only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the form, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0038] See also Figure 1 , a pituitary tumor dynamic risk assessment and monitoring method, comprising the following steps:

[0039] S1: Obtain dynamic monitoring data of pituitary tumors, including MRI imaging data and hormone level data;

[0040] The MRI image data includes MRI image data of a pituitary tumor and peripheral arterial blood vessels; the peripheral arterial blood vessels include an internal carotid artery and an anterior cerebral artery.

[0041] MRI (magnetic resonance imaging) can accurately show the size, shape, location and invasiveness of pituitary tumors. The present application determines the dynamic risk of pituitary tumors based on the size, shape and other characteristics of pituitary tumors. Pituitary tumors may compress or wrap around the surrounding arterial blood vessels. The internal carotid artery and the anterior cerebral artery are the arteries closest to the location of the pituitary tumor. The growth trend of the pituitary tumor is partially reflected in the hemodynamic characteristics of the artery. As the pituitary tumor grows, compression or wrapping of the artery will cause vascular stenosis or occlusion, thereby forming arterial plaques. The formation of such plaques may lead to insufficient blood supply to the brain, causing symptoms such as headaches, dizziness, and blurred vision. Therefore, the present invention combines the internal carotid artery and the anterior cerebral artery as important indicators for pituitary tumor risk assessment and monitoring.

[0042] The hormone level data include serum prolactin data, growth hormone data and thyroid hormone data.

[0043] S2: data processing and feature extraction, including MRI image data processing and feature extraction and hormone level data processing and feature extraction, to obtain MRI image data features and hormone level data features;

[0044] S21: MRI image data processing and feature extraction; including:

[0045] S211: MRI image data preprocessing, including: image data denoising and image segmentation. Optionally, the present invention performs image segmentation processing based on a Unet model combined with a Transformer to obtain the pituitary tumor body and the internal carotid artery and anterior cerebral artery image regions of interest.

[0046] S212: MRI image data feature extraction;

[0047] The features include pituitary tumor image features and blood flow information features. The pituitary tumor image features include size, shape, and position. The blood flow information features include hemodynamic characteristics obtained based on the internal carotid artery and the anterior cerebral artery, including blood flow, blood flow resistance, blood volume, and vascular morphology.

[0048] S22: Hormone level data processing and feature extraction;

[0049] Cluster analysis of hormone level data was performed based on statistical methods, including:

[0050] S221: Based on the improved FCM clustering algorithm, blood samples are clustered according to hormone level data and the samples are divided into C categories; the clustering algorithm is as follows:

[0051] 1. Initialize the membership matrix based on random membership value calculation;

[0052] 2. Calculate the first cluster center according to the following formula : ;in, N is the sample size, The fuzziness is m Time j Samples Belong to i The degree of class, The fuzziness is m Time k The samples belong to i The degree of class, m For fuzziness.

[0053] 3. Calculate the second cluster center according to the following formula : ;in, , Respectivelyk samples and t samples, For the k The samples belong to i The degree of class, For the t The samples belong to i The degree of class, C is the number of sample clusters;

[0054] 4. Combine the second cluster center And the membership value , update the Lagrange multiplier using the following formula : ;in, For the j samples; The fuzziness is m -1 hour j Samples Belong to i the degree of class;

[0055] 5. Combine the second cluster center and the Lagrange multiplier to calculate the updated membership matrix according to the following formula: ; Further obtain the updated membership matrix U.

[0056] 6. Repeat steps 2 to 5 until the change in the membership value in the latest two generations is less than the given threshold or the number of generations reaches the set value. Then stop the generation, defuzzify the latest membership matrix, and obtain the clustering result.

[0057] The clustering algorithm is calculated based on the cluster center and the remaining samples, so that the centroid stability of each sample cluster is improved, thereby improving the classification accuracy and stability.

[0058] S222: Based on C The sampling strategy is as follows:

[0059] The sampling step is divided into a first sampling step and a second sampling step. The first sampling step adopts unequal probability sampling without replacement; the second sampling step adopts simple random sampling.

[0060] S223: extracting features based on the sampled data;

[0061] The data characteristics of samples of different categories were obtained, including the mean, standard deviation, median of serum prolactin data, growth hormone data and thyroxine data, as well as the correlation between the three hormones.

[0062] S3: Establish a dynamic monitoring and prediction model;

[0063] The present invention performs dynamic monitoring and prediction of pituitary tumors based on the pre-trained BiGRU neural network attention model, which specifically includes the following:

[0064] The model is Figure 2 As shown, it includes a pre-training layer, a time-domain convolutional layer, a bidirectional gated recurrent network layer, and an attention mechanism layer;

[0065] The pre-training layer uses a pre-trained model to vectorize input features;

[0066] The time domain convolution layer introduces variable dilated convolution into the traditional convolution operation. That is, in the dilated convolution, by inserting gaps in the convolution kernel, the kernel's receptive field is expanded, and the network's ability to learn transient information of different lengths is enhanced, thereby improving the network's performance. The structure of the three-layer dilated convolution is shown in the figure below: Figure 3 As shown in Figure 2, the length of the filter is increased by skipping part of the input. represents input, and represents the output, and d represents the dilation rate, which reflects the size of the receptive field of the convolution kernel. As the number of network layers increases, the receptive field of the convolution kernel increases exponentially. Therefore, the dilated convolution can extract time series feature information from a farther time before the current moment. The receptive field size of the dilated convolution is: ;in, is the convolution kernel size, is the expansion rate, n is the number of network layers.

[0067] In addition, the temporal convolution layer consists of multiple stacked residual blocks. Each residual block contains two causal dilated convolution layers, a batch normalization layer, and an activation function layer. By introducing skip connections, some inputs are allowed to bypass the intermediate convolution operation. After the 1*1 convolution operation, it is directly added to the data that has undergone the intermediate convolution operation, which can effectively prevent problems such as gradient explosion when deep networks are stacked.

[0068] The bidirectional gated recurrent network layer combines the information features from the forward and backward directions, enabling it to fully learn the temporal features of the past and future. It consists of a GRU forward layer and a GRU backward layer, and the prediction value of the current moment is obtained by integrating the output results of the forward and backward layers. The calculation formulas of each layer are as follows: ;in, , They are t The state of the backward layer and the forward layer at all times, , They are t -1 moment backward and forward layer states, , are the weights corresponding to the backward and forward layers, Representation and time t The deviation corresponding to the hidden layer state at Output of the bidirectional gated recurrent network layer.

[0069] The weight coefficient of the attention mechanism layer is obtained by the following formula: ;in, is the output vector of the bidirectional gated recurrent network layer, is the weight, is the bias coefficient, is the randomly initialized attention mechanism matrix, is the attention weight matrix, is the intermediate coefficient, for The transpose of is the output of the attention mechanism layer.

[0070] S4: Calculate risk index;

[0071] Based on the MRI image data features and hormone level data features obtained above, the dynamic monitoring prediction model is input to obtain the pituitary tumor risk index.

[0072] In this embodiment, dynamic risk assessment and monitoring of pituitary tumors are realized, and a more objective and accurate dynamic risk prediction of pituitary tumors is achieved based on the pre-trained BiGRU neural network attention model and combined with MRI imaging data of pituitary tumors and peripheral arterial vessels and hormone level data; at the same time, blood samples are clustered according to hormone level data based on the improved FCM clustering algorithm, and combined with the stratified sampling method, representative data are selected for feature extraction, thereby improving the accuracy of subsequent predictions.

[0073] The embodiment of the present invention further provides a pituitary tumor dynamic risk assessment and monitoring system, based on the pituitary tumor dynamic risk assessment and monitoring method as described above, comprising:

[0074] A pituitary tumor dynamic monitoring data acquisition module is used to acquire pituitary tumor dynamic monitoring data, including MRI image data and hormone level data; the MRI image data includes MRI image data of the pituitary tumor and peripheral arterial blood vessels; the peripheral arterial blood vessels include the internal carotid artery and the anterior cerebral artery.

[0075] The data processing and feature extraction module is used for processing and feature extraction of MRI image data and hormone level data to obtain MRI image data features and hormone level data features; the MRI image data processing and features include pituitary tumor image features and blood flow information features, the pituitary tumor image features include size, shape, and position, and the blood flow information features include hemodynamic features obtained from the internal carotid artery and the anterior cerebral artery, including blood flow, blood flow resistance, blood volume, and vascular morphology. The hormone level data processing includes clustering blood samples according to hormone level data based on an improved FCM clustering algorithm.

[0076] A dynamic monitoring prediction model building module, which is used to dynamically monitor and predict pituitary tumors based on a pre-trained BiGRU neural network attention model;

[0077] The risk index calculation module is used to obtain the pituitary tumor risk index by inputting the dynamic monitoring prediction model based on the MRI image data characteristics and hormone level data characteristics obtained above.

[0078] In addition, an embodiment of the present invention also proposes a computer-readable storage medium, on which program instructions of a dynamic risk assessment and monitoring method for pituitary tumors are stored. The program instructions of the dynamic risk assessment and monitoring method for pituitary tumors can be executed by one or more processors to implement the steps of the dynamic risk assessment and monitoring method for pituitary tumors as described above.

[0079] The embodiments described above are only descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for dynamic risk assessment and monitoring of pituitary tumors, characterized in that: include: Obtain dynamic monitoring data of pituitary tumors, including MRI imaging data and hormone level data; Data processing and feature extraction, including MRI image data processing and feature extraction and hormone level data processing and feature extraction, to obtain MRI image data features and hormone level data features; The MRI image data processing and feature extraction includes pituitary tumor image feature extraction and blood flow information feature extraction, the pituitary tumor image features include size, shape, and position, and the blood flow information features include hemodynamic features obtained based on the internal carotid artery and the anterior cerebral artery, including blood flow, blood flow resistance, blood volume, and vascular morphology; The hormone level data processing and feature extraction specifically include: cluster analysis of hormone level data based on statistical methods, including: Based on the improved FCM clustering algorithm, blood samples were clustered according to hormone level data and the samples were divided into C categories; the clustering algorithm is as follows: Calculate the initial membership matrix based on the random membership values; The first cluster center is calculated according to the following formula : ； in, N is the sample size, The fuzziness is m Time j Samples Belong to i The degree of class, The fuzziness is m Time k The samples belong to i The degree of class, m is the fuzziness; The second cluster center is calculated according to the following formula : ； in, , Respectively k samples and t samples, For the k The samples belong to i The degree of class, For the t The samples belong to i The degree of class, C is the number of sample clusters; Combined with the second cluster center And the membership value , update the Lagrange multiplier using the following formula : ； in, For the j samples; The fuzziness is m -1 hour j Samples Belong to i the degree of class; Combine the second cluster center and the Lagrange multiplier, and calculate the updated membership matrix according to the following formula. The matrix elements are It is expressed as follows: ； Further obtain an updated membership matrix U; Repeat the steps from calculating the first cluster center to updating the membership matrix until the change in the membership value in the latest two generations is less than a given threshold or the number of generations reaches a set value, stop the generation, defuzzify the latest membership matrix, and obtain the clustering result; based on C The sampling strategy is as follows: The sampling step is divided into a first sampling step and a second sampling step, wherein the first sampling step adopts unequal probability sampling without replacement; the second sampling step adopts simple random sampling; Feature extraction based on sampled data; The data characteristics of samples of different categories were obtained, including the mean, standard deviation, median and correlation between the three hormones of serum prolactin, growth hormone and thyroxine data; Establish a dynamic monitoring prediction model to dynamically monitor and predict pituitary tumors based on the pre-trained BiGRU neural network attention model; The risk index was calculated and input into the dynamic monitoring prediction model based on the MRI image data characteristics and hormone level data characteristics obtained above to obtain the pituitary tumor risk index.

2. The pituitary tumor dynamic risk assessment and monitoring method according to claim 1, characterized in that: The MRI image data includes MRI image data of a pituitary tumor and peripheral arterial blood vessels; the peripheral arterial blood vessels include an internal carotid artery and an anterior cerebral artery.

3. The pituitary tumor dynamic risk assessment and monitoring method according to claim 2, characterized in that: The pre-trained BiGRU neural network attention model includes a pre-training layer, a time domain convolution layer, a bidirectional gated recurrent network layer, and an attention mechanism layer.

4. The pituitary tumor dynamic risk assessment and monitoring method according to claim 3, characterized in that: The time domain convolution layer introduces variable dilated convolution.

5. The pituitary tumor dynamic risk assessment and monitoring method according to claim 4, characterized in that: The time domain convolution layer is composed of multiple stacked residual blocks, each residual block contains two causal dilated convolution layers, a batch normalization layer and an activation function layer.

6. A pituitary tumor dynamic risk assessment and monitoring system, based on the pituitary tumor dynamic risk assessment and monitoring method according to any one of claims 1 to 5, characterized in that: include: A pituitary tumor dynamic monitoring data acquisition module, which is used to acquire pituitary tumor dynamic monitoring data, including MRI image data and hormone level data; the MRI image data includes MRI image data of the pituitary tumor and peripheral arterial blood vessels; the peripheral arterial blood vessels include the internal carotid artery and the anterior cerebral artery; A data processing and feature extraction module, which is used to process and extract features of MRI image data and hormone level data to obtain MRI image data features and hormone level data features; The MRI image data processing and features include pituitary tumor image features and blood flow information features, wherein the pituitary tumor image features include size, shape, and location, and the blood flow information features include hemodynamic features obtained based on the internal carotid artery and the anterior cerebral artery, including blood flow, blood flow resistance, blood volume, and vascular morphology; the hormone level data processing includes clustering blood samples based on hormone level data based on an improved FCM clustering algorithm; A dynamic monitoring prediction model building module, which is used to dynamically monitor and predict pituitary tumors based on a pre-trained BiGRU neural network attention model; The risk index calculation module is used to obtain the pituitary tumor risk index by inputting the dynamic monitoring prediction model based on the MRI image data characteristics and hormone level data characteristics obtained above.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions of a method for dynamic risk assessment and monitoring of pituitary tumors, and the program instructions of the method for dynamic risk assessment and monitoring of pituitary tumors can be executed by one or more processors to implement the steps of the method for dynamic risk assessment and monitoring of pituitary tumors as described in any one of claims 1 to 5.

Images