Interpretable AI method and system integrating neural network and nuclear magnetic resonance technology

Abstract

The invention provides an interpretable AI method and system integrating a neural network and a nuclear magnetic resonance technology, and relates to the technical field of medical image analysis. The method comprises the steps that image data of the brain and body parts of a patient are obtained through the nuclear magnetic resonance technology, statistical analysis of gray value distribution is carried out, the image data are converted into standardized gray value distribution within a target range, and standardized image data are obtained. According to the method, the multi-dimensional features are extracted layer by layer through the neural network, so that key details such as edges and textures in the image can be comprehensively captured, more accurate feature expression is realized, the image quality is improved after the deep learning model simulates and scores the image quality and optimizes scanning parameters, and the balance between the imaging efficiency and the quality is ensured; by providing higher interpretability for feature classification and distribution annotation of the optimized image, the applicability and efficiency of the nuclear magnetic resonance technology in the field of medical image analysis are improved, and the accuracy and intelligence of disease judgment are promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image analysis, and particularly to an interpretable AI method and system integrating neural network and nuclear magnetic resonance technology. Background Art

[0002] The technical field of medical image analysis includes a variety of techniques for detecting, judging, and evaluating the state of the internal structure and function of the human body. The core content of this field includes different imaging techniques such as X-ray, CT scan, magnetic resonance imaging (MRI), and ultrasonic imaging. By obtaining detailed images of the human body internal, it provides necessary data support for medical professionals to make disease judgments and monitor. The overall technical field is constantly developing, including improvements in hardware devices for image acquisition, innovations in image processing and analysis algorithms, and clinical application research on image results.

[0003] Among them, the interpretable AI method of neural network and nuclear magnetic resonance technology refers to the application of artificial intelligence, especially deep learning technology, to the analysis of MRI images. This method combines the detailed image data obtained by nuclear magnetic resonance imaging with neural network algorithms to automatically identify and extract key features in the images. Through the training and learning of a batch of MRI image data, the neural network can identify the pathological patterns in the images and judge the classification and severity of diseases. This technology mainly involves data processing and analysis, as well as the application of algorithms, to improve the accuracy and efficiency of medical judgment.

[0004] In the prior art, there is a lack of effective standardization means for the gray value distribution of nuclear magnetic resonance image data, resulting in deviations in the subsequent analysis of image data obtained under different devices and environmental conditions, affecting the accuracy and consistency of the results. The extraction of image features depends on a single algorithm or manual annotation, making it difficult to achieve multi-dimensional and multi-level feature expression, especially insufficient in the extraction of edge features and texture features, resulting in the omission of key lesion features. The optimization of the existing image quality more depends on fixed parameter settings and cannot be flexibly adjusted according to different individuals or different scanning conditions, thereby affecting the imaging effect and judgment efficiency. Traditional methods lack sufficient interpretability in image classification and annotation, unable to clearly show the understanding and judgment logic of the model for image features, reducing the trust of clinicians in the judgment results. These deficiencies limit the application potential of nuclear magnetic resonance technology in the analysis of complex lesions and personalized medicine, and are difficult to meet the needs of the development of precision medicine. Summary of the Invention

[0005] In order to solve the technical problems that the image data obtained under different devices and environmental conditions in the prior art are prone to deviation in subsequent analysis, affecting the accuracy and consistency of the results; the extraction of image features depends on a single algorithm or manual annotation, making it difficult to achieve multi-dimensional and multi-level feature expression, especially insufficient in the extraction of edge features and texture features, resulting in the omission of key lesion features; the optimization of the existing image quality depends more on fixed parameter settings and cannot be flexibly adjusted according to different individuals or different scanning conditions, thus affecting the imaging effect and judgment efficiency; the traditional methods lack sufficient interpretability in image classification and annotation, cannot clearly show the understanding and judgment logic of the model for image features, and reduce the trust of clinicians in the judgment results. The deficiencies limit the application potential of nuclear magnetic resonance technology in the analysis of complex lesions and personalized medicine and are difficult to meet the needs of the development of precision medicine. An interpretable AI method and system integrating neural network and nuclear magnetic resonance technology are provided in the embodiments of the present invention. The technical solutions are as follows:

[0006] On the one hand, an interpretable AI method integrating neural network and nuclear magnetic resonance technology is provided. The method is implemented by an interpretable AI device and includes:

[0007] S1: Obtain the image data of the patient's brain and body parts through nuclear magnetic resonance technology, perform statistical analysis on the gray value distribution, convert the image data into a standardized gray value distribution within the target range, and obtain standardized image data.

[0008] S2: Based on the standardized image data, use the neural network to layer by layer extract the edge features, texture features and multi-dimensional features of the image, identify the feature mapping through the weight matrix of the neural network, and perform a non-linear transformation on the feature mapping to obtain feature vector data.

[0009] S3: Apply a deep learning model to the feature vector data, simulate the impact of the adjustment of scanning parameters on the image quality according to the data distribution, score the image quality by comparing the scanning parameter combinations of different differential simulation results, and obtain optimized scanning parameters.

[0010] S4: Use the optimized scanning parameters to adjust the magnetic field intensity fluctuation range and frequency distribution of the nuclear magnetic resonance imaging device, optimize the repetition time and echo time of the pulse sequence, control the signal reception sensitivity within the scanning time, and obtain optimized image data.

[0011] S5: Use AI technology to perform transfer learning on the optimized image data, retain the edge features and texture features extracted from the neural network, classify the magnetic resonance imaging images by adjusting the weight matrix, and interpret the feature distribution and annotation criteria to obtain the disease type judgment result.

[0012] Optionally, the normalized image data includes multi-tissue images with noise interference removed, images with uniform gray-scale distribution, and images with optimized contrast.

[0013] The eigenvector data includes multi-dimensional space feature representations, lesion area feature distributions, and non-linear mapping feature sets extracted by neural networks.

[0014] The optimized scanning parameters include magnetic field parameters for optimizing the signal-to-noise ratio, time parameters for optimizing the image resolution, and parameter combinations for balancing the scanning speed and image quality.

[0015] The optimized image data includes structural imaging data with enhanced contrast, lesion images, and time-series images for capturing dynamic changes.

[0016] The disease type judgment results include disease judgment types based on classification models, location markings of lesion areas, and visual interpretation information of feature distributions.

[0017] Optionally, S1 includes:

[0018] S101: Obtain image data of the patient's brain and body parts through nuclear magnetic resonance technology, adjust the scanning intensity and sampling time of the nuclear magnetic resonance equipment to control the acquisition quality, monitor the image generation process in real time, eliminate noise and artifacts generated during the acquisition process, and obtain the original data set.

[0019] S102: Based on the original data set, perform gray-scale segmentation, separate noise regions, perform boundary correction on the gray-scale values of pixel points, eliminate brightness imbalance, and optimize the image hierarchy through multiple iterations to obtain the processed image data.

[0020] S103: Extract the gray-scale values of pixel points from the processed image data, map the gray-scale values to the standardized interval within the target range through normalization operations, analyze the pixel distribution, and eliminate abnormal gray-scale values to obtain the normalized image data.

[0021] Optionally, S2 includes:

[0022] S201: Use the normalized image data, utilize neural networks, decompose the edge regions of the image layer by layer, identify the gradient information of edge pixels, calculate the edge intensity values of the image, and perform connection operations on the edge regions to obtain edge feature records.

[0023] S202: Through the edge feature records, gradually extract the texture regions in the image, analyze the gray-scale distribution of texture pixel points, and analyze the texture direction and periodic features in combination with the spatial arrangement relationship to obtain the texture feature data.

[0024] S203: Use the texture feature data to perform a combined analysis of the texture information and edge features. Through non-linear transformation operations, extract the key feature components and obtain the feature vector data.

[0025] Optionally, the formula for calculating the edge intensity value of the image is as follows:

[0026] (1)

[0027] Where, represents the edge intensity value of the image; represents the gradient value of the image in the direction, represents the gradient value of the image in the direction, and are adjustment coefficients.

[0028] Optionally, S3 includes:

[0029] S301: According to the feature vector data, perform a distribution analysis of the feature components, simulate the image generation process under different scanning parameter conditions, and obtain the simulated image data set.

[0030] S302: Based on the simulated image data set, gradually compare the distribution of the image feature vectors generated by different scanning parameter combinations, calculate the mean, variance, and deviation degree of the feature components, and obtain the image quality score record.

[0031] S303: Through the image quality score record, screen the scanning parameter combinations with higher scores, and combine the score distribution to analyze the weights of the feature components under the simulated conditions to obtain the optimized scanning parameters.

[0032] Optionally, S4 includes:

[0033] S401: Based on the optimized scanning parameters, adjust the magnetic field intensity fluctuation range of the nuclear magnetic resonance imaging device, set the frequency distribution range of the magnetic field change, and perform tests by gradually adjusting the current intensity of the scanning coil to obtain the magnetic field optimization parameters.

[0034] S402: Use the magnetic field optimization parameters to optimize the repetition time and echo time of the pulse sequence, calculate the optimized parameter values, adjust the signal sampling frequency using the pulse interval within the scanning time, and evaluate the integrity and stability of the signal reception to obtain the signal reception record.

[0035] S403: Use the signal reception record to control the signal reception sensitivity within the scanning time, and obtain the optimized image data by adjusting the gain amplitude and signal filtering range of the signal reception.

[0036] Optionally, the formula for calculating the optimized parameter values is as follows:

[0037] (2)

[0038] Among them, is the optimized parameter value, represents the repetition time, represents the echo time, represents the signal sampling frequency, represents the pulse interval, , , are adjustment coefficients.

[0039] Optionally, S5 includes:

[0040] S501: Extract the edge features and texture features of the early layers of the neural network from the optimized image data. By freezing the weights of the early layers, evaluate the stability of feature extraction, and obtain the edge and texture feature data.

[0041] S502: Use the edge and texture feature data to adjust the weight matrix of the last few layers of the neural network, calculate the correlation between the classification target and the feature components, and optimize the classification boundary through multiple iterations to obtain the classification result of the magnetic resonance image.

[0042] S503: Based on the classification result of the magnetic resonance image, analyze the category correspondence relationship of the feature distribution, combine the disease annotation standard to interpret the classification result, extract the medical feature information corresponding to the classification label, and determine the disease category to obtain the disease type judgment result.

[0043] On the other hand, an interpretable AI system integrating neural network and nuclear magnetic resonance technology is provided. This system is applied to the interpretable AI method integrating neural network and nuclear magnetic resonance technology. The system includes:

[0044] An image acquisition module, used to acquire image data of the brain and body parts from a nuclear magnetic resonance imaging device, and convert the original data into a standardized gray value range by adjusting the gray value distribution to obtain standardized image data.

[0045] A feature mapping module, used to extract edge, texture, and multi-dimensional features layer by layer according to the standardized image data, and process the features by applying a weight matrix and a non-linear transformation to obtain feature vector data.

[0046] An impact evaluation module, used to use the feature vector data for a deep learning model, simulate the impact of scan parameter changes on image quality, and obtain optimized scan parameters through the analysis and comparison of differential simulation results.

[0047] The magnetic field intensity adjustment module is used to adjust the magnetic field intensity and frequency of the nuclear magnetic resonance equipment by using optimized scanning parameters, optimize the pulse sequence time, control the signal reception sensitivity, and obtain optimized image data.

[0048] The disease classification and interpretation module is used to perform transfer learning on the optimized image data, retain the edge and texture features extracted initially, classify the image by adjusting the weights of the deep network, interpret the feature distribution, and obtain the disease type judgment result.

[0049] On the other hand, there is provided an interpretable AI device, which includes: a processor; a memory, and computer-readable instructions are stored on the memory. When the computer-readable instructions are executed by the processor, any one of the interpretable AI methods integrating the neural network and nuclear magnetic resonance technology as described above is implemented.

[0050] On the other hand, there is provided a computer-readable storage medium, and at least one instruction is stored in the storage medium. The at least one instruction is loaded and executed by the processor to implement any one of the interpretable AI methods integrating the neural network and nuclear magnetic resonance technology as described above.

[0051] The beneficial effects brought by the technical solutions provided in the embodiments of the present invention at least include:

[0052] In the present invention, by statistically analyzing and standardizing the gray value distribution of nuclear magnetic resonance images, the consistency and comparability of imaging data are ensured, the image data deviation caused by equipment differences or environmental impacts is effectively reduced, and the accuracy of image analysis is improved. Combining the multi-dimensional features extracted layer by layer by the neural network can comprehensively capture key details such as edges and textures in the image and achieve more accurate feature expression. The deep learning model simulates and scores the image quality. After optimizing the scanning parameters, not only the image quality is improved, but also the interference of invalid scanning parameter settings on the imaging process is significantly reduced, ensuring the balance between imaging efficiency and quality. By dynamically adjusting the parameters of the imaging equipment, including elements such as magnetic field intensity, pulse sequence, and scanning time, the signal-to-noise ratio and resolution of nuclear magnetic resonance imaging are further improved, making the optimized image data have higher clinical applicability. The feature classification and distribution annotation of the optimized image provide higher interpretability. By retaining the early features in the classification process by the neural network, the reliability of the analysis and the accuracy of disease discrimination are improved. At the same time, the time cost and data requirements for model training are reduced, the applicability and efficiency of nuclear magnetic resonance technology in the field of medical image analysis are improved, and the precision and intelligence of disease judgment are promoted. Description of the Drawings

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the accompanying drawings required for description in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0054] Figure 1 It is a flowchart of an interpretable AI method integrating neural network and nuclear magnetic resonance technology provided by an embodiment of the present invention;

[0055] Figure 2 It is a refined flowchart of S1 of the present invention;

[0056] Figure 3 It is a refined flowchart of S2 of the present invention;

[0057] Figure 4 It is a refined flowchart of S3 of the present invention;

[0058] Figure 5 It is a refined flowchart of S4 of the present invention;

[0059] Figure 6 It is a refined flowchart of S5 of the present invention;

[0060] Figure 7 It is a block diagram of an interpretable AI system integrating neural network and nuclear magnetic resonance technology provided by an embodiment of the present invention;

[0061] Figure 8 It is a schematic structural diagram of an interpretable AI device provided by an embodiment of the present invention. Detailed implementation manners

[0062] The following describes the technical solutions in the present invention with reference to the accompanying drawings.

[0063] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to give examples, illustrations or explanations. Any embodiment or design solution described as "example" in the present invention should not be interpreted as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of the word "example" aims to present concepts in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two can be selected.

[0064] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same. "of", "corresponding", and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same.

[0065] In the embodiments of the present invention, sometimes subscripts such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meanings they express are the same.

[0066] To make the technical problems, technical solutions, and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0067] The embodiments of the present invention provide an interpretable AI method integrating neural networks and nuclear magnetic resonance technology. This method can be implemented by an interpretable AI device, which can be a terminal or a server. As Figure 1 shown in the flowchart of the interpretable AI method integrating neural networks and nuclear magnetic resonance technology, the processing flow of this method can include the following steps:

[0068] S1: Obtain image data of the patient's brain and body parts through nuclear magnetic resonance technology, perform statistical analysis on the gray value distribution of the original image, perform linear normalization processing according to the range and dynamic range of the image gray distribution, and convert the image data into a standardized gray value distribution within the target range to obtain standardized image data.

[0069] Among them, the standardized image data includes multi-tissue images with noise interference removed, images with uniform gray distribution, and images with optimized contrast.

[0070] Optionally, please refer to Figure 2 , the above step S1 may include the following steps S101 - S103:

[0071] S101: Obtain image data of the patient's brain and body parts through nuclear magnetic resonance technology, adjust the scanning intensity and sampling time of the nuclear magnetic resonance device to control the acquisition quality, and monitor the image generation process in real time to eliminate noise and artifacts generated during the acquisition process to obtain the original data set.

[0072] In a feasible implementation, magnetic resonance imaging (MRI) is a technology that uses strong magnetic fields and radio waves to image the body. This technology can produce high-resolution images and is very effective in judging various diseases such as cancer, heart disease, and musculoskeletal diseases. Adjusting the scanning intensity and sampling time is a key step in optimizing image quality. Appropriate adjustments can reduce scanning time and enhance image clarity. Real-time monitoring monitors data quality during the scanning process, adjusts equipment parameters in a timely manner, and eliminates noise and artifacts caused by patient movement or equipment problems. This series of operations ensures the accuracy and efficiency of data acquisition, enabling doctors and researchers to obtain reliable data for further analysis and judgment and obtain original data sets.

[0073] S102: Based on the original data set, grayscale segmentation is performed to separate the noise area, and the boundary correction of the grayscale value of the pixel point is performed to eliminate the brightness imbalance. The image hierarchy is optimized by combining multiple iterations to obtain the processed image data.

[0074] In a feasible implementation, the quality and usability of the image are improved. Grayscale segmentation mainly divides the grayscale value of the image into multiple segments to better analyze and process areas of different brightness. The separation of noise areas is to identify and isolate random noise in the image through an algorithm to ensure the clarity of the image. The boundary correction of pixel points involves adjusting the grayscale values between adjacent pixels to eliminate the brightness imbalance of the image, which involves complex mathematical calculations and image processing algorithms. Multiple iterative optimization processing is to perform multiple rounds of fine-tuning on the image. Each iteration improves a certain part or partial attributes of the image to obtain processed image data.

[0075] S103: extracting the grayscale value of the pixel from the processed image data, mapping the grayscale value to a standardized interval within the target range through a normalization operation, analyzing the pixel distribution and removing abnormal grayscale values to obtain standardized image data.

[0076] In a feasible implementation, normalization is an important data processing technology that simplifies the subsequent analysis and processing process by adjusting the data scale. Especially in the field of image processing, normalization can effectively unify the brightness and contrast standards of different images, so that the image analysis algorithm can more accurately identify and process image features. Pixel distribution analysis is to statistically analyze the gray value distribution of each pixel in the image, identify abnormal or unexpected gray values, such as overly dim or bright areas, areas related to image quality problems or data corruption. After eliminating outliers, the image can be further clarified, improving its value for clinical judgment and scientific research, facilitating storage, transmission and further image analysis, and obtaining standardized image data.

[0077] S2: Based on the standardized image data, use a neural network to extract the edge features, texture features, and multi-dimensional features of the image layer by layer. Identify the feature mapping through the weight matrix of each layer of the neural network and the input data, and perform a non-linear transformation on the feature mapping to obtain the feature vector data.

[0078] Among them, the feature vector data includes the multi-dimensional space feature representation, the lesion area feature distribution, and the non-linear mapping feature set extracted by the neural network.

[0079] Optionally, please refer to Figure 3 , the above step S2 may include the following steps S201 - S203:

[0080] S201: Adopt the standardized image data, use a neural network to decompose the edge area of the image layer by layer, identify the gradient information of the edge pixels, calculate the edge intensity value of the image, and perform a connection operation on the edge area to obtain the edge feature record.

[0081] Optionally, the formula for calculating the edge intensity value of the image is as follows:

[0082] (1)

[0083] Among them, represents the edge intensity value of the image; represents the gradient value of the image in the direction, represents the gradient value of the image in the direction, and are adjustment coefficients.

[0084] Parameter meaning and setting value:

[0085] and are calculated through the Sobel operator, representing the gradient values of the image in the and directions respectively. Set in a section of image data, the calculation result is 120, the calculation result is 80;

[0086] and are adjustment coefficients. According to the different image contents, the coefficients can be adjusted to optimize the edge detection effect. Through multiple experiments, it is determined that and can provide better results in various image scenarios. The values are obtained by monitoring the edge detection effects of different types of images (such as urban, natural, and indoor scenes) and optimizing the parameters;

[0087] Substitute the parameters into the formula for calculation:

[0088] ;

[0089] The results show that the edge intensity of the image at this point is relatively high, indicating that the edge features are relatively obvious. The numerical results are used for further threshold processing to determine which parts of the edges should be retained as significant features of the image. This kind of calculation provides a quantitative basis for image edge detection and relies on the edge intensity values to construct a clearer image edge map.

[0090] S202: Through edge feature recording, gradually extract the texture regions in the image, analyze the gray-scale distribution of texture pixel points, and combine the spatial arrangement relationship to analyze the texture direction and periodic characteristics to obtain texture feature data.

[0091] In a feasible implementation, edge feature recording is to identify the significant boundary lines in the image through image processing techniques. The boundary lines mark the boundaries between different objects or texture regions. Gradually extracting the texture regions in the image is to separate the regions with different texture characteristics according to the edge lines. Analyzing the gray-scale distribution of texture pixel points involves calculating the gray-scale values of the pixels in each texture region and evaluating the uniformity or variability of the distribution, which helps to identify the texture patterns in the image. Analyzing the spatial arrangement relationship further reveals the directionality and periodicity of the texture, such as the ripples or repeating patterns of the texture. Features are the key to understanding the surface characteristics of objects. The texture direction can indicate the flow direction of the object surface, while the periodic characteristics can reflect the repeatability of the structure, which is crucial for subsequent image analysis and machine vision applications, such as applications in materials science, biomedicine, and remote sensing, to obtain texture feature data.

[0092] S203: Use the texture feature data to perform combined analysis on the texture information and edge features, and through non-linear transformation operations, extract the key feature components to obtain feature vector data.

[0093] In a feasible implementation, the texture feature data provides detailed information about the surface of an object in an image, and the combined analysis combines the texture information with the edge features of the image to provide a more comprehensive image analysis. The non-linear transformation operation is a key step in the feature extraction process, which involves applying mathematical models such as neural networks or deep learning algorithms to identify and extract non-linear relationships in the image data. The relationships contain complex features that are crucial for image description. The key feature components are the most representative features extracted from the batch data, which are particularly effective for classification, recognition, or image processing tasks. The feature vector data is the mathematical representation of the key features and can be used in machine learning models for further image classification or recognition. The successful execution of this process depends on accurate mathematical models and efficient algorithms to ensure that useful information can be extracted from complex image data and converted into an operable numerical form, applicable to various practical application scenarios such as the vision system of autonomous vehicles, security monitoring, and applications in medical image analysis, etc., to obtain the feature vector data.

[0094] S3: Apply a deep learning model to the feature vector data, calculate the critical weights of the features according to the data distribution, simulate the impact of adjusting the scanning parameters on the image quality, score the image quality by comparing the parameter combinations of the differential simulation results, and obtain the optimized scanning parameters.

[0095] Among them, the optimized scanning parameters include the magnetic field parameters for optimizing the signal-to-noise ratio, the time parameters for optimizing the image resolution, and the parameter combination for balancing the scanning speed and the image quality.

[0096] Optionally, please refer to Figure 4 , the above step S3 may include the following steps S301 - S303:

[0097] S301: According to the feature vector data, conduct a distribution analysis on the feature components, simulate the image generation process under different scanning parameters, and obtain a simulated image data set.

[0098] In a feasible implementation, through in-depth analysis of the feature vector data, where the feature vectors contain the key information of the image data and the data is sourced from previous image processing and feature extraction steps, the distribution analysis of the feature components includes statistically analyzing the frequency, central tendency, and dispersion degree of each feature component. The statistical data helps to understand the role and manifestation form of the feature components in the image. The simulation of different scanning parameters is achieved by changing the scanning settings of the nuclear magnetic resonance imaging device, such as the magnetic field strength, frequency distribution, and scanning coil current intensity, etc., to observe how the changes affect the generation of the image. This simulation helps to optimize the parameter settings during the actual scanning process to obtain the best image quality. Conducted in a controlled environment, it not only verifies the accuracy of the feature component distribution analysis but also provides an experimental platform to predict and verify the specific impact of different scanning parameters on the image quality. This process is of great significance for the development and application of medical imaging technology, allowing technicians to understand and optimize the imaging parameters before actual application, improving the judgment value of the image and reducing the scanning risk for patients, and obtaining a simulated image dataset.

[0099] S302: Based on the simulated image dataset, gradually compare the distribution of the image feature vectors generated by different scanning parameter combinations, calculate the mean value, variance, and deviation degree of the feature components, and obtain the image quality score record.

[0100] In a feasible implementation, compare the distribution of the image feature vectors generated by different scanning parameter combinations, according to the formula: and calculate the mean value and variance of the feature components. In the formula, represents the mean value of the feature components, represents the variance of the feature components, represents the feature component of the th feature vector, represents the total number of feature vectors;

[0101] By monitoring and collecting the feature vector data , the mean value of all feature vectors can be calculated. By calculating the sum of the squares of the differences between the feature components of each feature vector and the mean value and then taking the average, the variance is obtained;

[0102] Suppose the feature component values of 5 feature vectors are [2, 3, 5, 4, 6] respectively, calculate the mean value:

[0103] ;

[0104] Calculate the variance:

[0105] ;

[0106] This process is used to analyze the subtle differences in how different scanning parameters affect the feature vector distribution.

[0107] S303: Through the image quality score records, screen out the scanning parameter combinations with higher scores. Combining the score distribution, analyze the weights of the feature components under simulated conditions to obtain the optimized scanning parameters.

[0108] In a feasible implementation, by analyzing the score records, the weights of the feature components under simulated conditions are analyzed, which depends on the in-depth analysis of statistical data such as the mean and variance calculated in the previous step. Connect the statistical data with the image quality scores, compare the image quality under different parameter combinations in detail. The comparison results support the optimized decision of parameter selection, guide future image scanning operations to ensure high-quality image output in practical applications. Through precise parameter adjustment and optimization, the overall quality of the image can be significantly improved, which is particularly important for fields that require high-precision image analysis. The optimization process also includes the analysis of the weights of the feature components, which is a key step in determining which features play a decisive role in image quality assessment, can adjust and set the scanning parameters more accurately, and achieve the optimal image capture effect in various application scenarios to obtain the optimized scanning parameters.

[0109] S4: Use the optimized scanning parameters to adjust the magnetic field intensity fluctuation range and frequency distribution of the nuclear magnetic resonance imaging device, optimize the repetition time and echo time of the pulse sequence, control the signal reception sensitivity during the scanning time, and re-execute data acquisition according to the adjusted parameter configuration to obtain the optimized image data.

[0110] Among them, the optimized image data includes structure imaging data with enhanced contrast, lesion images, and time series images for capturing dynamic changes.

[0111] Optionally, please refer to Figure 5 The above step S4 may include the following steps S401 - S403:

[0112] S401: Based on the optimized scanning parameters, adjust the magnetic field intensity fluctuation range of the nuclear magnetic resonance imaging device, set the frequency distribution range of the magnetic field change, and conduct tests by gradually adjusting the current intensity of the scanning coil to obtain the magnetic field optimization parameters.

[0113] In a feasible implementation, in magnetic resonance imaging (MRI) technology, precise control of the magnetic field is the key to obtaining high-quality images. Optimizing the scanning parameters includes adjusting the magnetic field strength and setting the frequency distribution. These parameters directly affect the signal quality and image resolution during the imaging process. Adjusting the fluctuation range of the magnetic field strength is to ensure that under different scanning requirements, the magnetic field can provide sufficient strength to excite the resonance of hydrogen atoms in the body. The setting of the frequency distribution is related to how the magnetic field effectively covers the entire scanning area. By precisely controlling the current intensity in the scanning coil, technicians can finely adjust the performance of the magnetic field to meet the specific needs of different cases. Gradually adjusting and testing the parameters can help medical physicists and technicians determine the optimal scanning settings, maximizing the image quality while ensuring safety, enhancing not only the imaging efficiency but also the accuracy of judgment, and providing better medical services for patients to obtain the magnetic field optimization parameters.

[0114] S402: Utilize the magnetic field optimization parameters to optimize the repetition time and echo time of the pulse sequence, calculate the optimized parameter values, adjust the signal sampling frequency using the pulse interval within the scanning time, evaluate the integrity and stability of signal reception, and obtain the signal reception record.

[0115] Optionally, the formula for calculating the optimized parameter values is as follows:

[0116] (2)

[0117] Where, is the optimized parameter value, represents the repetition time, represents the echo time, represents the signal sampling frequency, represents the pulse interval, 、 、 are adjustment coefficients.

[0118] Parameter meanings and setting values:

[0119] (repetition time) is obtained from the actual operation parameters of the MRI device, set to 2000 milliseconds, which determines the time interval between each repeated emission of the pulse sequence;

[0120] (echo time) is also read from the MRI device parameters, set to 30 milliseconds, which affects the delay degree of the signal echo;

[0121] (signal sampling frequency) is set according to the actual MRI sampling requirements, set to 500 Hz, which directly affects the speed and quality of data acquisition;

[0122] (Pulse interval) is obtained from the actual pulse sequence setting, set to 15 milliseconds, which affects the time layout of the pulse sequence; the weight coefficients 、 、 are set to 0.6, 0.4 and 0.02 respectively. The values are obtained based on past empirical data and experimental tuning. The selection of weight coefficients takes into account the different degrees of influence of each parameter on signal stability;

[0123] Substitute the parameters into the formula for calculation:

[0124]

[0125] The results show that by precisely adjusting the scanning parameters of MRI and optimizing the parameter process, the reception quality and stability of the signal can be significantly improved. This value reflects that the optimized parameter settings can effectively improve the performance of MRI scanning, ensure high-quality judgment images, and directly affect the quality of MRI images and the accuracy of judgment.

[0126] S403: Adopt signal reception recording to control the signal reception sensitivity during the scanning time, and obtain optimized image data by adjusting the gain amplitude of signal reception and the signal filtering range.

[0127] In a feasible implementation, signal reception recording is a crucial step in magnetic resonance imaging. This process involves capturing the signals excited by the scanning coil and converting them into image data. Controlling the signal reception sensitivity during the scanning time is to accurately capture the weak signals from inside the human body. Adjusting the gain amplitude of signal reception is related to the amplification degree of the signal, which needs to be adjusted according to specific imaging requirements and the characteristics of the target tissue to ensure signal quality without generating excessive noise. Adjusting the signal filtering range is to filter out unnecessary signal components, reduce image interference and improve imaging clarity. Optimized operations make the details more accurate, provide higher support for disease judgment, and improve the application value of MRI technology in clinical practice to obtain optimized image data.

[0128] S5: Use AI technology to perform transfer learning on the optimized image data, retain the edge features and texture features extracted from the earlier layers of the neural network, classify the magnetic resonance imaging images by adjusting the weight matrix of the last few layers, and interpret the feature distribution and annotation criteria to obtain the disease type judgment result.

[0129] Among them, the disease type judgment result includes the disease judgment type based on the classification model, the localization annotation of the lesion area, and the visual interpretation information of the feature distribution.

[0130] Optionally, please refer to Figure 6, the above step S5 may include the following steps S501 - S503:

[0131] S501: Extract the edge features and texture features of the early layers of the neural network from the optimized image data. By freezing the weights of the early layers, evaluate the stability of feature extraction and obtain the edge and texture feature data.

[0132] In a feasible implementation, extract the edge and texture features from the optimized image data according to the formulas: and Calculate the edge and texture feature values of the image. In the formula, represents the edge feature value of the image, represents the texture feature value of the image, represents the pixel value of the image at position ; represents the pixel value of the image at position ; represents the pixel value of the image at position ;

[0133] For a given optimized image data set, traverse each pixel point, calculate the absolute value of the difference from the adjacent pixel points on the right and below, and accumulate the values respectively to obtain the measures of the edge and texture features;

[0134] Set the pixel values of a 3x3 image area as [[1, 2, 3], [4, 5, 6], [7, 8, 9]], then the edge feature value is calculated as:

[0135] ;

[0136] The texture feature value is calculated as:

[0137] ;

[0138] The results show that it is possible to quantify the saliency of edges and textures in the image, providing a data basis for subsequent neural network weight freezing and feature stability evaluation.

[0139] S502: Use the edge and texture feature data to adjust the weight matrix of the last few layers of the neural network, calculate the correlation between the classification target and the feature components, and obtain the magnetic resonance image classification result by optimizing the classification boundary through multiple iterations.

[0140] In a feasible implementation, the weight matrices of the last few layers of the neural network are adjusted, the correlation between the classification target and the feature components is calculated, and the classification boundary is optimized through multiple iterations. This process involves in-depth understanding and adjustment of the neural network structure to ensure that the network can correctly interpret and utilize the input feature data. Through the iterative process, the network weights are continuously adjusted to ensure that each step of optimization is based on the results of the previous step, gradually reducing the error and improving the classification accuracy, reflecting how the neural network can effectively distinguish different categories by learning the given features, not only improving the classification accuracy but also increasing the sensitivity to different lesion features, and obtaining the magnetic resonance image classification results.

[0141] S503: Based on the magnetic resonance image classification results, analyze the category correspondence of the feature distribution, interpret the classification results in combination with the disease annotation standard, extract the medical feature information corresponding to the classification label, and determine the disease category to obtain the disease type judgment result.

[0142] In a feasible implementation, analyze the category correspondence of the feature distribution, interpret the classification results in combination with the disease annotation standard. The key to this process is to combine the output of the machine learning model with medical expertise to provide interpretable judgment information. By extracting the medical feature information corresponding to the classification label, doctors can better understand the judgment basis of the model and further determine the disease category. This not only increases the transparency of the classification results but also improves the reliability of medical judgments, provides important decision-making information for clinical practice, makes disease management more precise and personalized. This integrated analysis process provides an efficient tool for modern medicine, helps to quickly and accurately identify and classify various disease states, and obtains the disease type judgment result.

[0143] In the embodiments of the present invention, by statistically analyzing and standardizing the gray value distribution of nuclear magnetic resonance images, the consistency and comparability of imaging data are ensured, the deviation of image data caused by equipment differences or environmental impacts is effectively reduced, and the accuracy of image analysis is improved. Combining the multi-dimensional features extracted layer by layer by the neural network can comprehensively capture key details such as edges and textures in the image and achieve more accurate feature expression. The deep learning model simulates and scores the image quality. After optimizing the scanning parameters, not only the image quality is improved, but also the interference of invalid scanning parameter settings on the imaging process is significantly reduced, ensuring the balance between imaging efficiency and quality. By dynamically adjusting the parameters of the imaging equipment, including elements such as magnetic field strength, pulse sequence, and scanning time, the signal-to-noise ratio and resolution of nuclear magnetic resonance imaging are further improved, making the optimized image data have higher clinical applicability. Providing higher interpretability for the feature classification and distribution annotation of the optimized images, through the early features retained by the neural network during the classification process, improving the reliability of analysis and the accuracy of disease discrimination, while reducing the time cost and data requirements of model training, enhancing the applicability and efficiency of nuclear magnetic resonance technology in the field of medical image analysis, and promoting the precision and intelligence of disease judgment.

[0144] Figure 7 FIG. is a block diagram of an interpretable AI system integrating a neural network and nuclear magnetic resonance technology according to an exemplary embodiment. This system is used for an interpretable AI method integrating a neural network and nuclear magnetic resonance technology. Referring to Figure 7 , this system includes an image acquisition module, a feature mapping module, an impact assessment module, a magnetic field strength adjustment module, and a disease classification and interpretation module. Among them:

[0145] The image acquisition module is used to acquire image data of the brain and body parts from a nuclear magnetic resonance imaging device. By adjusting the gray value distribution, the original data is converted into a standardized gray value range to obtain standardized image data.

[0146] The feature mapping module is used to extract edge, texture, and multi-dimensional features layer by layer according to the standardized image data, and process the features using a weight matrix and a non-linear transformation to obtain feature vector data.

[0147] The impact assessment module is used to use the feature vector data for a deep learning model to simulate the impact of changes in scanning parameters on image quality. By analyzing and comparing the differential simulation results, optimized scanning parameters are obtained.

[0148] The magnetic field strength adjustment module is used to use the optimized scanning parameters to adjust the magnetic field strength and frequency of the nuclear magnetic resonance device, optimize the pulse sequence time, control the signal reception sensitivity, and obtain optimized image data.

[0149] A disease classification and interpretation module for performing transfer learning on the optimized image data, retaining the edge and texture features extracted initially, classifying the images by adjusting the weights of the deep network, interpreting the feature distribution, and obtaining the disease type judgment result.

[0150] In the embodiment of the present invention, by statistically analyzing and standardizing the gray value distribution of nuclear magnetic resonance images, the consistency and comparability of the imaging data are ensured, the image data deviation caused by equipment differences or environmental impacts is effectively reduced, and the accuracy of image analysis is improved. Combining the multi-dimensional features extracted layer by layer by the neural network can comprehensively capture key details such as edges and textures in the image, achieving a more accurate feature expression. The deep learning model simulates and scores the image quality. After optimizing the scanning parameters, not only the image quality is improved, but also the interference of invalid scanning parameter settings on the imaging process is significantly reduced, ensuring the balance between imaging efficiency and quality. By dynamically adjusting the parameters of the imaging equipment, including elements such as magnetic field strength, pulse sequence, and scanning time, the signal-to-noise ratio and resolution of nuclear magnetic resonance imaging are further improved, making the optimized image data have higher clinical applicability. The feature classification and distribution annotation of the optimized image provide higher interpretability. By retaining the early features in the classification process by the neural network, the reliability of the analysis and the accuracy of disease discrimination are improved. At the same time, the time cost and data requirements for model training are reduced, enhancing the applicability and efficiency of nuclear magnetic resonance technology in the field of medical image analysis and promoting the precision and intelligence of disease judgment.

[0151] Figure 8 It is a schematic structural diagram of an interpretable AI device provided by an embodiment of the present invention. As Figure 8 shown, the interpretable AI device may include the interpretable AI system integrating the above-mentioned Figure 7 shown integrated neural network and nuclear magnetic resonance technology. Optionally, the interpretable AI device 410 may include a first processor 2001.

[0152] Optionally, the interpretable AI device 410 may further include a memory 2002 and a transceiver 2003.

[0153] Among them, the first processor 2001, the memory 2002, and the transceiver 2003, such as can be connected through a communication bus.

[0154] Next, in combination with Figure 8 each component of the interpretable AI device 410 will be specifically introduced:

[0155] Among them, the first processor 2001 is the control center of the interpretability AI device 410, which can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 is one or more central processing units (CPUs), or can be an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention, such as: one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs).

[0156] Optionally, the first processor 2001 can execute various functions of the interpretability AI device 410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0157] In a specific implementation, as an embodiment, the first processor 2001 can include one or more CPUs, such as Figure 8 the CPU0 and CPU1 shown in

[0158] In a specific implementation, as an embodiment, the interpretability AI device 410 can also include multiple processors, such as Figure 8 the first processor 2001 and the second processor 2004 shown in. Each of these processors can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). Here, the processor can refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions).

[0159] Among them, the memory 2002 is used to store the software program for executing the solution of the present invention and is controlled by the first processor 2001 for execution. The specific implementation manner can refer to the above method embodiments and will not be elaborated here.

[0160] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or may also be an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently and is coupled to the first processor 2001 through an interface circuit ( Figure 8 not shown) of the interpretable AI device 410. The embodiments of the present invention do not make specific limitations on this.

[0161] The transceiver 2003 is used to communicate with a network device or communicate with a terminal device.

[0162] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 8 not shown separately). Among them, the receiver is used to implement the receiving function, and the transmitter is used to implement the sending function.

[0163] Optionally, the transceiver 2003 may be integrated with the first processor 2001 or may exist independently and is coupled to the first processor 2001 through an interface circuit ( Figure 8 not shown) of the interpretable AI device 410. The embodiments of the present invention do not make specific limitations on this.

[0164] It should be noted that Figure 8 the structure of the interpretable AI device 410 shown does not constitute a limitation on the router. The actual knowledge structure recognition device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0165] In addition, the technical effects of the interpretable AI device 410 can refer to the technical effects of the interpretable AI method integrating neural networks and nuclear magnetic resonance technology described in the above method embodiments and will not be elaborated here.

[0166] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0167] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM) or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of random access memory (RAM) are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced SDRAM (ESDRAM), synchlink DRAM (SLDRAM) and direct rambus RAM (DR RAM).

[0168] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0169] It should be understood that the term "and / or" in this document is merely a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. Additionally, the character " / " in this document generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood by referring to the context before and after.

[0170] In the present invention, "at least one" means one or more, and "a plurality" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single-item (individual) or plural-item (individual). For example, at least one of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple.

[0171] It should be understood that in various embodiments of the present invention, the magnitudes of the sequence numbers of the above processes do not mean the order of execution is prior or subsequent. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0172] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention.

[0173] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described devices, apparatuses, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0174] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling, direct coupling, or communication connection can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.

[0175] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0176] In addition, the functional units in each embodiment of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0177] When the above-described functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0178] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An interpretable AI method integrating neural network and nuclear magnetic resonance technology, characterized in that, The method comprises: S1: Obtain image data of the patient's brain and body parts through nuclear magnetic resonance technology, perform statistical analysis on gray value distribution, convert the image data into standardized gray value distribution within the target range, and obtain standardized image data; S2: Based on the standardized image data, using a neural network, extracting edge features, texture features, and multi-dimensional features of the image layer by layer, identifying feature mapping through a weight matrix of the neural network, and performing nonlinear transformation on the feature mapping to obtain feature vector data; S3: applying a deep learning model to the feature vector data, simulating the effect of adjusting the scanning parameters on the image quality according to the data distribution, scoring the image quality by comparing the scanning parameter combinations of the differentiated simulation results, and obtaining the optimized scanning parameters; S4: using the optimized scanning parameters, adjusting the magnetic field intensity fluctuation range and frequency distribution of the nuclear magnetic resonance imaging device, optimizing the repetition time and echo time of the pulse sequence, controlling the signal receiving sensitivity within the scanning time, and acquiring optimized image data; S5: Use AI technology to perform transfer learning on the optimized image data, retain the edge features and texture features extracted from the neural network, classify the magnetic resonance imaging images by adjusting the weight matrix, interpret the feature distribution and labeling standards, and obtain the disease type judgment result.

2. The interpretable AI method integrating a neural network and nuclear magnetic resonance technology according to claim 1, wherein The standardized image data includes a multi-tissue image with noise interference eliminated, a grayscale uniform distribution image, and a contrast optimized image; The feature vector data includes multi-dimensional spatial feature representation, lesion area feature distribution and non-linear mapping feature set extracted by neural network; The optimized scanning parameters include magnetic field parameters for optimizing signal-to-noise ratio, time parameters for optimizing image resolution, and a parameter combination for balancing scanning speed and image quality; The optimized image data include contrast-enhanced structural imaging data, lesion images, and time series images captured by dynamic changes; The disease type judgment result includes the disease judgment type based on the classification model, the positioning annotation of the lesion area and the visual explanation information of the feature distribution.

3. The interpretable AI method integrating a neural network and nuclear magnetic resonance technology according to claim 1, characterized in that, Said S1 comprises: S101: Obtain image data of the patient's brain and body parts through nuclear magnetic resonance technology, adjust the scanning intensity and sampling time of the nuclear magnetic resonance equipment to control the acquisition quality, monitor the image generation process in real time, eliminate the noise and artifacts generated during the acquisition process, and obtain the original data set; S102: Based on the original data set, grayscale segmentation is performed to separate the noise area, and boundary correction of the grayscale value of the pixel point is performed to eliminate the brightness imbalance, and the image hierarchy is optimized by combining multiple iterations to obtain processed image data; S103: extracting the grayscale value of the pixel from the processed image data, mapping the grayscale value to a standardized interval within a target range through a normalization operation, analyzing the pixel distribution and removing abnormal grayscale values to obtain standardized image data.

4. The interpretable AI method integrating a neural network and nuclear magnetic resonance technology according to claim 1, wherein The S2 comprises: S201: Using the standardized image data, utilize a neural network to decompose the edge region of the image layer by layer, identify the gradient information of edge pixels, calculate the edge intensity value of the image, perform a connection operation on the edge region, and obtain an edge feature record; S202: Through the edge feature record, gradually extract the texture region in the image, analyze the gray-scale distribution of texture pixel points, and analyze the texture direction and periodic characteristics in combination with the spatial arrangement relationship to obtain texture feature data; S203: Utilize the texture feature data to perform a combined analysis of texture information and edge features, and through a non-linear conversion operation, extract key feature components to obtain feature vector data.

5. The interpretable AI method integrating a neural network and nuclear magnetic resonance technology according to claim 4, wherein The formula for calculating the edge intensity value of the image is as follows: （1） Among them, represents the edge intensity value of the image; represents the gradient value of the image in the direction, represents the gradient value of the image in the direction, and are adjustment coefficients.

6. The interpretable AI method integrating a neural network and nuclear magnetic resonance technology according to claim 1, wherein The S3 includes: S301: According to the feature vector data, perform a distribution analysis on the feature components, simulate the image generation process under different scanning parameter combinations, and obtain a simulated image data set; S302: Based on the simulated image data set, gradually compare the image feature vector distributions generated by different scanning parameter combinations, calculate the mean value, variance, and deviation degree of the feature components, and obtain an image quality score record; S303: Through the image quality score record, screen the scanning parameter combinations with higher scores, and analyze the weights of the feature components under the simulated conditions in combination with the score distribution to obtain optimized scanning parameters.

7. The interpretable AI method integrating a neural network and nuclear magnetic resonance technology according to claim 1, characterized in that, The S4 includes: S401: Based on the optimized scanning parameters, adjust the magnetic field intensity fluctuation range of the nuclear magnetic resonance imaging device, set the frequency distribution range of magnetic field changes, and perform tests by gradually adjusting the current intensity of the scanning coil to obtain magnetic field optimization parameters; S402: Utilize the magnetic field optimization parameters to optimize the repetition time and echo time of the pulse sequence, calculate the optimized parameter values, adjust the signal sampling frequency using the pulse interval within the scanning time, and evaluate the integrity and stability of signal reception to obtain a signal reception record; S403: Adopt the signal reception record to control the signal reception sensitivity during the scanning time, and obtain optimized image data by adjusting the gain amplitude and signal filtering range of signal reception.

8. The interpretable AI method integrating a neural network and nuclear magnetic resonance technology according to claim 7, characterized in that, The formula for calculating the optimized parameter values is as follows: （2） Among them, is the optimized parameter value, represents the repetition time, represents the echo time, represents the signal sampling frequency, represents the pulse interval, , , are adjustment coefficients.

9. The interpretable AI method integrating a neural network and nuclear magnetic resonance technology according to claim 1, characterized in that, The S5 includes: S501: Through the optimized image data, extract the edge features and texture features of the early layers of the neural network, evaluate the stability of feature extraction by freezing the weights of the early layers, and obtain edge and texture feature data; S502: Utilize the edge and texture feature data to adjust the weight matrix of the latter several layers of the neural network, calculate the correlation between the classification target and the feature components, and optimize the classification boundary through multiple iterations to obtain a magnetic resonance image classification result; S503: Based on the magnetic resonance image classification result, analyze the category correspondence relationship of the feature distribution, interpret the classification result in combination with the disease annotation standard, extract the medical feature information corresponding to the classification label, and determine the disease category to obtain a disease type judgment result.

10. An interpretable AI system integrating neural networks and nuclear magnetic resonance technology, the interpretable AI system integrating neural networks and nuclear magnetic resonance technology is used to implement the interpretable AI method integrating neural networks and nuclear magnetic resonance technology as described in any one of claims 1-9, characterized in that, The system includes: An image acquisition module, which is used to acquire image data of the brain and body parts from a nuclear magnetic resonance imaging device, convert the original data into a standardized gray value range by adjusting the gray value distribution, and obtain standardized image data; A feature mapping module, which is used to extract edge, texture and multi-dimensional features layer by layer according to the standardized image data, process the features by applying a weight matrix and a non-linear transformation, and obtain feature vector data; An impact assessment module, which is used to use the feature vector data for a deep learning model, simulate the impact of scanning parameter changes on image quality, and obtain optimized scanning parameters through the analysis and comparison of the differential simulation results; A magnetic field intensity adjustment module, which is used to adjust the magnetic field intensity and frequency of the nuclear magnetic resonance device by using the optimized scanning parameters, optimize the pulse sequence time, control the signal reception sensitivity, and obtain optimized image data; A disease classification and interpretation module, which is used to perform transfer learning on the optimized image data, retain the edge and texture features extracted initially, classify the image by adjusting the weights of the deep network, interpret the feature distribution, and obtain the disease type judgment result.

Images