Cerebrovascular analysis method and system based on ultrasonic data and large language model

Abstract

The invention discloses a cerebrovascular analysis method based on ultrasonic data and a large language model, and the method comprises the steps: carrying out the processing and coding of cerebrovascular ultrasonic data obtained through transcranial ultrasonic equipment, and obtaining a structured training data set; training a Mama model by using the structured training data set to construct a cerebrovascular analysis model; encoding cerebrovascular data acquired in real time through transcranial ultrasonic equipment, inputting the data into the cerebrovascular analysis model, and outputting cerebrovascular structure parameters and cerebrovascular function parameters; and generating an analysis report according to the cerebrovascular structure parameters, the cerebrovascular function parameters and cerebrovascular data collected in real time. Therefore, cerebrovascular state parameters and functional parameters can be accurately analyzed, a standardized analysis report can be generated, a full-automatic process from data acquisition and analysis to report generation is realized, the cerebrovascular state and function can be accurately analyzed, errors caused by manual operation are reduced, and the accuracy of analysis is improved. And the accuracy and reliability of medical image analysis can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer-aided diagnosis, and particularly to a cerebrovascular analysis method, system and computer-readable storage medium based on ultrasonic data and large language models. Background Art

[0002] Stroke is the second leading cause of death globally and has a high disability rate, seriously threatening human health. The Circle of Willis (CoW), as the core structure for cerebral blood flow compensation regulation, its integrity directly affects the blood supply stability of brain tissue. Research shows that only about 20% of the population has a complete CoW structure, and structural variations may lead to a decline in cerebral blood flow reserve capacity. Transcranial Doppler ultrasound has become a common tool for detecting the hemodynamic parameters of the CoW due to its non-invasive, real-time, low-cost and other advantages.

[0003] However, affected by the skull acoustic window limitation and the complexity of vascular structures, the transcranial Doppler ultrasound detection process depends on the operator's experience, and the detection and analysis processes are very time-consuming; the operator's interpretation of images has a certain subjectivity, affecting the consistency of parameter interpretation.

[0004] Based on this, a new solution is needed. Summary of the Invention

[0005] The main purpose of the present invention is to provide a cerebrovascular analysis method and system based on ultrasonic data and large language models.

[0006] To achieve the above purpose, the present invention provides a cerebrovascular analysis method based on ultrasonic data and large language models, including the following steps:

[0007] Process and encode the cerebrovascular ultrasonic data obtained by a transcranial ultrasound device to obtain a structured training data set, where the cerebrovascular ultrasonic data includes ultrasonic image data and corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data;

[0008] Use the structured training data set to train a Mamba model to construct a cerebrovascular analysis model;

[0009] Encode and process the cerebrovascular data collected in real time by a transcranial ultrasound device and input it into the cerebrovascular analysis model to output cerebrovascular structure parameters and cerebrovascular function parameters; and

[0010] Generate an analysis report according to the cerebrovascular structure parameters, the cerebrovascular function parameters, and the cerebrovascular data collected in real time.

[0011] In the cerebrovascular analysis method based on ultrasonic data and large language models provided by the present invention, the steps of processing and encoding the cerebrovascular ultrasonic data obtained by a transcranial ultrasonic device to obtain a structured training data set include:

[0012] Analyze the ultrasonic image data and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain video data analysis results;

[0013] Encode the video data analysis results and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain analysis embedding encoding vectors;

[0014] Embed the analysis embedding encoding vectors into the ultrasonic image data to obtain the structured training data set.

[0015] In the cerebrovascular analysis method based on ultrasonic data and large language models provided by the present invention, the steps of encoding the video data analysis results and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain analysis embedding encoding vectors include:

[0016] Perform normalization processing on the blood flow velocity data to obtain a normalization result;

[0017] Convert the blood flow direction data into one-hot encoding;

[0018] Perform fast Fourier transform on the waveform data and the spectral analysis data, extract spectral features, and convert the spectral features into spectral vectors of a fixed length;

[0019] Combine the one-hot encoding, the normalization result, the spectral feature vectors, and the video data analysis results into the analysis embedding encoding vectors.

[0020] In the cerebrovascular analysis method based on ultrasonic data and large language models provided by the present invention, the steps of embedding the analysis embedding encoding vectors into the ultrasonic image data to obtain the structured training data set include:

[0021] Convert the ultrasonic image data into a plurality of two-dimensional image blocks;

[0022] Linearly project the plurality of two-dimensional image blocks into a vector space, and add the analysis embedding encoding vectors and class labels related to the cerebrovascular structure parameters and the cerebrovascular function parameters to obtain the structured training data set.

[0023] In the cerebrovascular analysis method based on ultrasonic data and large language models provided by the present invention, the Mamba model includes a plurality of stacked Mamba modules. The Mamba module includes a main processing module and a skip connection module. The main processing module includes an input linear projection unit, a convolutional unit, a SiLU activation function unit, a forward selective state space model processing unit, a backward selective state space model processing unit, and an output linear projection unit. The skip connection module includes a SiLU activation function unit. The output of the Mamba module is the residual connection between the output of the main processing module and the output of the skip connection module.

[0024] In addition, to achieve the above object, the present invention also provides a cerebrovascular analysis system based on ultrasonic data and large language models, including:

[0025] A structured training dataset generation module, configured to process and encode cerebrovascular ultrasonic data obtained by a transcranial ultrasonic device to obtain a structured training dataset. The cerebrovascular ultrasonic data includes ultrasonic image data and corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data;

[0026] A model training module, configured to train a Mamba model using the structured training dataset to construct a cerebrovascular analysis model;

[0027] An analysis module, configured to encode and process cerebrovascular data collected in real time by a transcranial ultrasonic device and input the processed data into the cerebrovascular analysis model, and output cerebrovascular structure parameters and cerebrovascular function parameters; and

[0028] A report generation module, configured to generate an analysis report based on the cerebrovascular structure parameters, the cerebrovascular function parameters, and the cerebrovascular data collected in real time.

[0029] In the cerebrovascular analysis system based on ultrasonic data and large language models provided by the present invention, the structured training dataset generation module includes:

[0030] A feature extraction unit, configured to analyze the ultrasonic image data and corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain a video data analysis result;

[0031] A data encoding unit, configured to encode the video data analysis result and corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain an analysis embedding encoding vector;

[0032] A data merging unit, configured to embed the analysis embedding encoding vector into the ultrasonic image data to obtain the structured training dataset.

[0033] In the cerebrovascular analysis system based on ultrasonic data and large language model provided by the present invention, the data merging unit includes:

[0034] A data conversion sub-unit, configured to convert the ultrasonic image data into a plurality of two-dimensional image blocks;

[0035] A data generation sub-unit, configured to linearly project the plurality of two-dimensional image blocks into a vector space, and add the analysis embedding coding vector and class markers related to the cerebrovascular structure parameters and the cerebrovascular function parameters to obtain the structured training data set.

[0036] In the cerebrovascular analysis system based on ultrasonic data and large language model provided by the present invention, the Mamba model includes a plurality of stacked Mamba modules. The Mamba module includes a main processing module and a skip connection module. The main processing module includes an input linear projection unit, a convolution unit, a SiLU activation function unit, a forward selective state space model processing unit, a backward selective state space model processing unit, and an output linear projection unit. The skip connection module includes a SiLU activation function unit. The output of the Mamba module is the residual connection between the output of the main processing module and the output of the skip connection module.

[0037] The present invention also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned cerebrovascular analysis method based on ultrasonic data and large language model are implemented.

[0038] The cerebrovascular analysis system and method based on ultrasonic data and large language model provided by the present invention have the following beneficial effects: The cerebrovascular analysis method based on ultrasonic data and large language model provided by the present invention uses a structured data set encoded by ultrasonic image data and corresponding blood flow velocity data, blood flow direction data, waveform data, and spectrum analysis data to train a Mamba model to construct a cerebrovascular analysis model, which can accurately analyze cerebrovascular state parameters and function parameters, and generate a standardized analysis report. Thus, a fully automated process from data acquisition, analysis to report generation is realized, and accurate analysis of cerebrovascular state and function can be achieved, reducing errors caused by manual operations, and helping to improve the accuracy and reliability of medical image analysis. Description of the Drawings

[0039] 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 drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts:

[0040] Figure 1 The following is a flowchart of a cerebrovascular analysis method based on ultrasonic data and a large language model provided by an embodiment of the present invention. Detailed implementation manners

[0041] To facilitate the understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Typical embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, these embodiments are provided to make the disclosure of the present invention more thorough and comprehensive.

[0042] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0043] Figure 1 The following is a flowchart of a cerebrovascular analysis method based on ultrasonic data and a large language model provided by an embodiment of the present invention. As Figure 1 shown, the cerebrovascular analysis method based on ultrasonic data and a large language model includes the following steps:

[0044] Step S1: Process and encode the cerebrovascular ultrasonic data obtained by a transcranial ultrasonic device to obtain a structured training data set;

[0045] Specifically, in an embodiment of the present invention, the cerebrovascular ultrasonic data includes ultrasonic image data and corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data. First, a large amount of transcranial Doppler ultrasonic image video data (each frame of the image can be regarded as a sample in the sequence) and / or relevant information data of cerebral blood flow (including blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data) are obtained, and denoising, contrast enhancement, and normalization processing are performed on them to ensure the consistency and quality of the data. Then, the obtained cerebrovascular ultrasonic data is analyzed to obtain video data analysis results, which include but are not limited to whether cerebral blood vessels appear and their relative positions in the image and their internal blood flow states, etc. Among them, cerebral blood vessels include but are not limited to: ipsilateral / contralateral middle cerebral artery M1 segment, ipsilateral / contralateral anterior cerebral artery A1 segment, anterior cerebral artery A2 segment, ipsilateral / contralateral posterior cerebral artery P1 segment, ipsilateral / contralateral posterior cerebral artery P2 segment, anterior communicating artery, ipsilateral / contralateral posterior communicating artery. Then, the video data analysis results and / or the relevant information data of cerebral blood flow are encoded together to obtain an analysis embedding coding vector. Finally, the analysis embedding coding vector is embedded into the ultrasonic image data to obtain a training data set. Therefore, step S1 includes:

[0046] Step S11: Analyze the ultrasonic image data and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain the video data analysis result;

[0047] Specifically, in this embodiment, various methods, such as but not limited to traditional methods and deep learning methods, are used to analyze the transcranial Doppler ultrasonic image data and / or the relevant information data of cerebral blood flow, and results such as whether cerebral blood vessels appear and their relative positions in the image are obtained as the video data analysis result, which are used as parameters for subsequent training of the model to control the projection weights of different image blocks, so that the model can pay more attention to important regions.

[0048] Step S12: Encode the video data analysis result and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain the analysis embedded coding vector;

[0049] Specifically, in this embodiment, the data to be sent into the model for training needs to be encoded and converted into an analysis embedded coding vector to facilitate subsequent model training. Therefore, step S12 includes:

[0050] Step S121: Perform normalization processing on the blood flow velocity data to obtain the normalization result;

[0051] In this embodiment, the blood flow velocity is processed by normalization encoding, and the blood flow velocity (range 0 - 200 cm / s) is mapped to the range [0, 1] using Min - Max normalization,

[0052]

[0053] where min_vel = 0 and max_vel = 200.

[0054] Step S122: Convert the blood flow direction data into one - hot encoding;

[0055] In this embodiment, one - hot encoding is performed on the blood flow direction, that is, the blood flow direction (towards or away from the probe) is converted into one - hot encoding. For example, towards the probe towards → [1, 0], away from the probe away → [0, 1].

[0056] Step S123: Perform fast Fourier transform on the waveform data and the spectral analysis data, extract spectral features, and convert the spectral features into a fixed - length spectral vector;

[0057] In this embodiment, fast Fourier transform (FFT) is performed on waveform and spectrum data, and important features in the spectrum, such as the main frequency, peak frequency, etc., are extracted. Then, the extracted spectrum features are converted into vectors of a fixed length. Thus, in the subsequent model establishment process, the analysis of vascular function and status can be realized from more dimensions.

[0058] Step S124: Combine the one-hot encoding, the normalization result, the spectrum feature vector, and the video data analysis result into the analysis embedding encoding vector E res 。

[0059] Step S13: Embed the analysis embedding encoding vector into the ultrasonic image data to obtain the structured training dataset.

[0060] Specifically, in an embodiment of the present invention, after encoding the video data analysis result and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectrum analysis data, it is necessary to integrate the encoded analysis embedding encoding vector E res with the corresponding 2D transcranial Doppler ultrasound image data. Specifically, first, the 2D transcranial Doppler ultrasound image data is converted into tiled 2D image blocks. Each image block can be regarded as a small segment, representing a part of the image. Then, these image blocks are linearly projected into a vector space, and the analysis embedding encoding E res is added. Therefore, step S13 includes:

[0061] Step S131: Convert the ultrasonic image data into a plurality of two-dimensional image blocks;

[0062] In this embodiment, the 2D image t ∈ R H×W×C is converted into tiled 2D image blocks where (H, W) represents the size of the input image, C represents the number of channels, and P represents the size of the image block.

[0063] Step S132: Linearly project the plurality of two-dimensional image blocks into a vector space, and add the analysis embedding encoding vector and class labels related to the cerebrovascular structure parameters and the cerebrovascular function parameters to obtain the structured training dataset.

[0064] In this embodiment, the image block x p is linearly projected to the size of the vector D, and the analysis embedding encoding is added

[0065] T0 = [t cls ; t1pW; t2pW; …; t J pW] + E res

[0066] where tj p is the j-th segment of the image patch t, which are learnable projection weights. In particular, a class token related to the cerebrovascular structure parameters and the cerebrovascular function parameters is also used to label the entire sequence, so that the model can better understand different important clinical diagnostic evidence related to transcranial Doppler and finally generate a corresponding report, denoted as t cls . The cerebrovascular structure parameters include but are not limited to whether a certain ipsilateral / contralateral branch vessel is detected in step S1, and the cerebrovascular function parameters include but are not limited to vessel diameter, vessel direction, peak systolic blood flow velocity, mean blood flow velocity, end-diastolic blood flow velocity, pulsatility index, and resistance index.

[0067] Through the above process, the deep features of the ultrasound image data can be effectively projected into the 1D sequence data, and the E res containing cerebral blood flow-related data and analysis results is embedded into the image data, providing a basis for subsequent model training and inference.

[0068] Step S2: Use the structured training dataset to train the Mamba model to construct a cerebrovascular analysis model.

[0069] Specifically, in an embodiment of the present invention, after data processing and encoding are completed, the encoded data is continuously fed into the Mamba model composed of stacked Mamba modules. Use codes such as T0, T1,..., Tn to represent each layer, where n is the number of layers of the model. In each layer, the data will be processed by the Mamba module, and each Mamba module will perform feature extraction and update on the data.

[0070] Specifically, in an embodiment of the present invention, the Mamba model includes a plurality of stacked Mamba modules. The Mamba module includes a main processing module and a skip connection module. The main processing module includes an input linear projection unit, a convolutional unit, a SiLU activation function unit, a forward selective state space model processing unit, a backward selective state space model processing unit, and an output linear projection unit. The skip connection module includes a SiLU activation function unit. The output of the Mamba module is the residual connection between the output of the main processing module and the output of the skip connection module.

[0071] During the training process, the data processed in step S1 is used to train and update the weight parameters of the neurons in the linear projection unit and the convolutional unit in each Mamba module, as well as the parameter matrices B, C, and the discretization step Δ in the subsequent state space model (SSM). The backpropagation algorithm is used to optimize the above weight parameters. The processing result T of each layer l will be used as the next layer T l+1input until the last layer T n 。

[0072] In this embodiment, the model parameters are optimized by the backpropagation algorithm so that the model can accurately analyze and evaluate the state of the cerebrovascular. During the training process, the weights of the Mamba module are continuously adjusted to better capture the features of the image data.

[0073] Further, in an embodiment of the present invention, a structured medical report example is used as training data to be input into the model at the same time so that the model can output a standard analysis report. The analysis report includes:

[0074] Overview of detection results: Describe the detected vascular structure parameters and functional parameters, for example, such as blood flow velocity values, vascular morphology, etc.;

[0075] Abnormal findings: Point out the undetected cerebrovascular or the abnormal parts, for example, "a certain branch vessel is not detected";

[0076] Evaluation of the integrity of the circle of Willis: Comprehensively analyze the overall structure of the circle of Willis, for example, "The structure of the circle of Willis is incomplete.

[0077] Step S3: Encode and process the cerebrovascular data collected in real time by the transcranial ultrasound device and input it into the cerebrovascular analysis model to output cerebrovascular structure parameters and cerebrovascular function parameters;

[0078] Specifically, in an embodiment of the present invention, during the state analysis and evaluation process, the cerebrovascular data collected in real time by the transcranial ultrasound device is encoded and processed according to the process of step S1, and then sent to the cerebrovascular analysis model established in step S2 for state analysis and evaluation to accurately evaluate the cerebrovascular structure parameters and functional parameters. Specifically, the cerebrovascular structure parameters include but are not limited to whether a certain ipsilateral / contralateral branch vessel is detected as described in step S1; the cerebrovascular function parameters include but are not limited to vessel diameter, vessel direction, peak systolic blood flow velocity, mean blood flow velocity, end-diastolic blood flow velocity, pulsatility index, resistance index.

[0079] Specifically, in an embodiment of the present invention, the encoded real-time data is sent to the main processing module and the skip connection module of the Mamba module. The main processing module first performs a linear projection on the input data, then passes through a convolutional layer and a SiLU activation function, then enters the selective state space model (Selective SSM) processing in both the forward and backward directions, and finally outputs through a linear projection. The data input to the skip connection module only passes through the SiLU activation function and then is added to the output of the main processing module to form a residual connection. Such a structure is repeated multiple times until the cerebrovascular structure parameters and cerebrovascular function parameters are output.

[0080] Among them, the basic process of the Selective State Space Model (Selective SSM) is as follows:

[0081] · Input: x: (B, L, D)

[0082] · Output: y: (B, L, D)

[0083] · Among them, B is the batch size, L is the expansion dimension, and D is the state space dimension

[0084] Steps:

[0085] 1. Define the parameter A: (D, N), representing a structured N×N matrix.

[0086] 2. Calculate B: (B, L, N): B = s B (x).

[0087] 3. Calculate C: (B, L, N): C = s C (x).

[0088] 4. Calculate Δ: (B, L, D): Δ = τ Δ (Parameter + s Δ (x)).

[0089] 5. Discretize A and B: A, B = discretize(Δ, A, B).

[0090] 6. Call SSM for time-varying recursive (scanning) processing: y ← SSM(A, B, C)(x).

[0091] 7. Return y.

[0092] During the inference process, the state analysis and evaluation process is completed through the following formula:

[0093] h′(t) = A·h(t) + B·x(t)

[0094] y(t) = C·h(t)

[0095] In particular, that is, a scalar multiplied by the identity matrix is used as the state matrix A to achieve a faster clinical inference speed.

[0096] The actual state analysis and evaluation process is the discretized form of the above formula. For the next element of the sequence, first determine the step size Δ required for discretization through the weights of the model, and then determine the position of t + 1 through Δ. In this way, first update the hidden state h(t + 1), and then update the final state y(t + 1).

[0097] Step S4: Generate an analysis report based on the cerebrovascular structure parameters, the cerebrovascular function parameters, and the real-time collected cerebrovascular data;

[0098] Specifically, after completing the status analysis and evaluation, an analysis report is generated according to the obtained cerebrovascular structure parameters, cerebrovascular function parameters, and real-time collected cerebrovascular data according to the model trained based on the standardized report template, so as to organize and present the information of the cerebrovascular structure parameters and function parameters analyzed and evaluated by the model as an analysis report that can be directly used by clinicians.

[0099] The cerebrovascular analysis method based on ultrasound data and large language model provided by the present invention can accurately analyze the cerebrovascular status parameters and function parameters and generate a standardized analysis report through the cerebrovascular analysis model constructed by training the Mamba model with the structured data set encoded from the ultrasound image data and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data. Thus, a fully automated process from data acquisition, analysis to report generation is realized, which can achieve accurate analysis of the cerebrovascular status and function, reduce the errors caused by manual operations, and help improve the accuracy and reliability of medical image analysis.

[0100] Correspondingly, the present invention also provides a cerebrovascular analysis system based on ultrasound data and large language model, including: a structured training data set generation module for processing and encoding the cerebrovascular ultrasound data obtained by a transcranial ultrasound device to obtain a structured training data set, where the cerebrovascular ultrasound data includes ultrasound image data and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data; a model training module for training the Mamba model with the structured training data set to construct a cerebrovascular analysis model; an analysis module for encoding and processing the real-time collected cerebrovascular data through the transcranial ultrasound device and inputting it into the cerebrovascular analysis model to output cerebrovascular structure parameters and cerebrovascular function parameters; and a report generation module for generating an analysis report based on the cerebrovascular structure parameters, the cerebrovascular function parameters, and the real-time collected cerebrovascular data.

[0101] Specifically, in an embodiment of the present invention, the structured training data set generation module includes: a feature extraction unit for analyzing the ultrasound image data and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain a video data analysis result; a data encoding unit for encoding the video data analysis result and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain an analysis embedding encoding vector; and a data merging unit for embedding the analysis embedding encoding vector into the ultrasound image data to obtain the structured training data set.

[0102] Specifically, in an embodiment of the present invention, the data merging unit includes: a data conversion sub-unit for converting the ultrasonic image data into a plurality of two-dimensional image blocks; a data generation sub-unit for linearly projecting the plurality of two-dimensional image blocks into a vector space, and adding the analysis embedding coding vector and class labels related to the cerebrovascular structure parameters and the cerebrovascular function parameters to obtain the structured training data set.

[0103] Specifically, in an embodiment of the present invention, the Mamba model includes a plurality of stacked Mamba modules. The Mamba module includes a main processing module and a skip connection module. The main processing module includes an input linear projection unit, a convolutional unit, a SiLU activation function unit, a forward selective state space model processing unit, a backward selective state space model processing unit, and an output linear projection unit. The skip connection module includes a SiLU activation function unit. The output of the Mamba module is the residual connection between the output of the main processing module and the output of the skip connection module.

[0104] An embodiment of the present invention further provides a cerebrovascular analysis device based on ultrasonic data and a large language model, which may include:

[0105] A memory for storing a computer program;

[0106] A processor, when executing the computer program stored in the above-mentioned memory, can implement the following steps:

[0107] Process and encode the cerebrovascular ultrasonic data obtained by a transcranial ultrasonic device to obtain a structured training data set. The cerebrovascular ultrasonic data includes ultrasonic image data and corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data; use the structured training data set to train the Mamba model to construct a cerebrovascular analysis model; encode and process the cerebrovascular data collected in real time by the transcranial ultrasonic device and input it into the cerebrovascular analysis model to output cerebrovascular structure parameters and cerebrovascular function parameters; and generate an analysis report according to the cerebrovascular structure parameters, the cerebrovascular function parameters, and the cerebrovascular data collected in real time.

[0108] An embodiment of the present invention further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the following steps can be implemented;

[0109] Process and encode cerebrovascular ultrasound data obtained by a transcranial ultrasound device to obtain a structured training dataset. The cerebrovascular ultrasound data includes ultrasound image data and corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data. Train a Mamba model using the structured training dataset to construct a cerebrovascular analysis model. Encode and process the cerebrovascular data collected in real time by the transcranial ultrasound device and input it into the cerebrovascular analysis model to output cerebrovascular structure parameters and cerebrovascular function parameters. And generate an analysis report based on the cerebrovascular structure parameters, the cerebrovascular function parameters, and the cerebrovascular data collected in real time.

[0110] The computer-readable storage medium may include: various media such as USB flash drives, external hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0111] In the specification provided herein, a large number of specific details are set forth. However, it is understood that embodiments of the present invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0112] Similarly, it should be understood that, in order to streamline the present disclosure and assist in understanding one or more of the various inventive aspects, in the foregoing description of the exemplary embodiments of the present invention, the various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the following claims, the inventive aspects lie in less than all the features of the preceding disclosed single embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of the present invention.

[0113] Those skilled in the art can understand that the modules in the devices in the embodiments can be adaptively changed and arranged in one or more devices different from those of the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and in addition, they can be divided into multiple sub-modules or sub-units or sub-components. Except that at least some of such features and / or processes or units are mutually exclusive, any combination can be adopted to combine all the features disclosed in this specification (including the accompanying claims, abstract and drawings) and all the processes or units of any method or device so disclosed. Unless otherwise clearly stated, each feature disclosed in this specification (including the accompanying claims, abstract and drawings) can be replaced by an alternative feature that provides the same, equivalent or similar purpose.

[0114] In addition, those skilled in the art can understand that although some of the embodiments herein include certain features included in other embodiments rather than other features, the combination of the features of different embodiments means that it is within the scope of the present invention and forms different embodiments. For example, in the following claims, any one of the claimed embodiments can be used in any combination.

[0115] Each component embodiment of the present invention can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components according to the embodiments of the present invention. The present invention can also be implemented as a device or device program (such as a computer program and a computer program product) for executing part or all of the methods described herein. Such a program for implementing the present invention can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0116] It should be noted that the above embodiments are illustrative of the present invention and not restrictive thereof, and alternative embodiments can be designed by those skilled in the art without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in the claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention can be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In a unit claim listing several devices, several of these devices can be embodied by the same item of hardware. The use of the words first, second, and third, etc. does not denote any order. These words can be interpreted as names.

Claims

1. A cerebrovascular analysis method based on ultrasonic data and large language models, characterized in that, It includes the following steps: Process and encode the cerebrovascular ultrasound data obtained by a transcranial ultrasound device to obtain a structured training dataset, where the cerebrovascular ultrasound data includes ultrasound image data and corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data; Use the structured training dataset to train a Mamba model to construct a cerebrovascular analysis model; Encode and process the cerebrovascular data collected in real time by a transcranial ultrasound device and input it into the cerebrovascular analysis model to output cerebrovascular structure parameters and cerebrovascular function parameters; And Generate an analysis report based on the cerebrovascular structure parameters, the cerebrovascular function parameters, and the cerebrovascular data collected in real time.

2. The cerebrovascular analysis method based on ultrasonic data and large language model according to claim 1, characterized in that The step of processing and encoding the cerebrovascular ultrasound data obtained by a transcranial ultrasound device to obtain a structured training dataset includes: Analyze the ultrasound image data and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain video data analysis results; Encode the video data analysis results and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain an analysis embedding encoding vector; Embed the analysis embedding encoding vector into the ultrasound image data to obtain the structured training dataset.

3. The cerebrovascular analysis method based on ultrasonic data and large language model according to claim 2, characterized in that, The step of encoding the video data analysis results and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain an analysis embedding encoding vector includes: Perform normalization processing on the blood flow velocity data to obtain a normalization result; Convert the blood flow direction data into a one-hot encoding; Perform a fast Fourier transform on the waveform data and the spectral analysis data, extract spectral features, and convert the spectral features into a fixed-length spectral vector; Combine the one-hot encoding, the normalization result, the spectral feature vector, and the video data analysis results into the analysis embedding encoding vector.

4. The cerebrovascular analysis method based on ultrasonic data and large language model according to claim 2, wherein, The step of embedding the analysis embedding encoding vector into the ultrasound image data to obtain the structured training dataset includes: Convert the ultrasound image data into multiple two-dimensional image blocks; Linearly project the multiple two-dimensional image blocks into a vector space, and add the analysis embedding encoding vector and class labels related to the cerebrovascular structure parameters and the cerebrovascular function parameters to obtain the structured training dataset.

5. The cerebrovascular analysis method based on ultrasonic data and large language models according to claim 1, characterized in that, The Mamba model includes a plurality of stacked Mamba modules. The Mamba module includes a main processing module and a skip connection module. The main processing module includes an input linear projection unit, a convolutional unit, a SiLU activation function unit, a forward selective state space model processing unit, a backward selective state space model processing unit, and an output linear projection unit. The skip connection module includes a SiLU activation function unit. The output of the Mamba module is the residual connection between the output of the main processing module and the output of the skip connection module.

6. A cerebrovascular analysis system based on ultrasonic data and large language models, characterized in that, It includes: A structured training dataset generation module for processing and encoding cerebrovascular ultrasound data obtained by a transcranial ultrasound device to obtain a structured training dataset, where the cerebrovascular ultrasound data includes ultrasound image data and corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data; A model training module for training a Mamba model using the structured training dataset to construct a cerebrovascular analysis model; An analysis module for encoding and processing cerebrovascular data collected in real time by a transcranial ultrasound device and inputting it into the cerebrovascular analysis model to output cerebrovascular structure parameters and cerebrovascular function parameters; And A report generation module for generating an analysis report based on the cerebrovascular structure parameters, the cerebrovascular function parameters, and the cerebrovascular data collected in real time.

7. The cerebrovascular analysis system based on ultrasonic data and large language model according to claim 6, characterized in that, The structured training dataset generation module includes: A feature extraction unit for analyzing the ultrasound image data and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain a video data analysis result; A data encoding unit for encoding the video data analysis result and the corresponding blood flow velocity data, blood flow direction data, waveform data, and spectral analysis data to obtain an analysis embedding encoding vector; A data merging unit for embedding the analysis embedding encoding vector into the ultrasound image data to obtain the structured training dataset.

8. The cerebrovascular analysis method based on ultrasonic data and large language model according to claim 7, characterized in that The data merging unit includes: A data conversion sub-unit for converting the ultrasound image data into a plurality of two-dimensional image blocks; A data generation sub-unit for linearly projecting the plurality of two-dimensional image blocks into a vector space and adding the analysis embedding encoding vector and class labels related to the cerebrovascular structure parameters and the cerebrovascular function parameters to obtain the structured training dataset.

9. The cerebrovascular analysis system based on ultrasonic data and large language model according to claim 6, wherein The Mamba model includes a plurality of stacked Mamba modules, and each Mamba module includes a main processing module and a skip connection module. The main processing module includes an input linear projection unit, a convolutional unit, a SiLU activation function unit, a forward selective state space model processing unit, a backward selective state space model processing unit, and an output linear projection unit. The skip connection module includes a SiLU activation function unit, and the output of the Mamba module is a residual connection between the output of the main processing module and the output of the skip connection module.

10. A computer-readable storage medium storing a computer program therein, characterized in that: When the computer program is executed by a processor, it implements the method according to any one of claims 1 to 5.

Images