Intelligent management method and system for diabetics based on artificial intelligence

Abstract

The invention relates to an intelligent management method and system for diabetics based on artificial intelligence. The method comprises the following steps: constructing a text data set and an image data set at a specified time point; carrying out text feature extraction and image feature extraction, and carrying out feature conversion on image feature data to obtain corresponding text features; performing feature fusion processing on the text features based on the text data and the text features based on the image data at the same time point according to a specified time point; creating and training a disease development prediction model, and inputting feature fusion data to predict the disease at discrete time points; acquiring the disease development condition of the whole time period and the disease development condition of each time period, and performing planning and diagnosis and treatment planning; and obtaining examination data after a specified period, predicting illness state data at the same time point by the illness state development prediction model, and if the illness state is more serious than the predicted illness state, adjusting diet and diagnosis and treatment plans. According to the process, the individuation of the diabetics and the intelligent management of the whole life cycle are realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent management of diabetic patients, and in particular relates to an intelligent management method and system for diabetic patients based on artificial intelligence. Background Art

[0002] Diabetes is a chronic disease characterized by hyperglycemia, caused by absolute or relative insulin deficiency and utilization disorders. The disease is mainly divided into three types: type 1, type 2 and gestational diabetes. The cause is mainly attributed to the combined effects of genetic and environmental factors, including decreased insulin secretion caused by islet cell dysfunction, or the body's insensitivity to insulin, or both, which prevents glucose in the blood from being effectively utilized and stored. Some diabetic patients and families have a clustering phenomenon of the disease. In addition, the incidence and prevalence of diabetes are on the rise worldwide.

[0003] When a healthy person develops type 1 or type 2 diabetes, the progression of diabetes is a one-way street for the patient, and it is irreversible. The patient can only slow down the progression of diabetes by controlling diet or exercising. When blood sugar is well controlled through daily diet or exercise, the progression of diabetes can be slowed down to a large extent. When blood sugar is poorly controlled due to an unreasonable daily diet or lack of exercise, the progression of diabetes may be accelerated.

[0004] Therefore, the daily management of diabetic patients is very important. Traditionally, the daily management of diabetic patients is carried out by listening to the doctor's advice. However, since diagnosis and treatment are only carried out at certain time points, the doctor's advice is only for the specified time point and cannot achieve intelligent management of diabetic patients throughout their life cycle. Therefore, how to provide personalized and intelligent management of diabetic patients throughout their life cycle is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent management method and system for diabetic patients based on artificial intelligence, so as to provide personalized and intelligent management of diabetic patients throughout their life cycle.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows: In a first aspect, a method for intelligent management of diabetic patients based on artificial intelligence is provided, comprising the following steps: S1: Obtaining personal information of the current diabetic patient and various examination report data at various time points, classifying the various examination report data into text data and image data, and preprocessing the text data and image data; S2: sorting the preprocessed text data according to the time points, and constructing a text data set at the specified time point; sorting the preprocessed image data according to the time points, and constructing an image data set corresponding to the specified time point; S3: extracting text features from the text data set at a specified time point, extracting image features from the image data at a specified time point, and performing feature conversion on the extracted image feature data to obtain text features based on the image data; S4: performing feature fusion processing on the text features based on the text data and the text features based on the image data at the same time point at the specified time point to obtain feature fusion data at each specified time point; S5: creating a disease progression prediction model and training it with the labeled data of diabetic patients, inputting the feature fusion data of each designated time point into the disease progression prediction model to predict the disease at discrete time points between the designated time points; S6: based on the disease prediction results of discrete time points between the specified time points, obtaining the disease progression of the current diabetic patient from the first diagnosis of diabetes to the current time point, and the disease progression of each time period obtained based on adjacent specified time points; S7: Create a diet planning model and train it based on the labeled diet planning data, and plan the diet for a specified time period starting from the current time point based on the disease progression in each time period, and plan the diagnosis and treatment for a specified time period starting from the current time point based on the disease progression in each time period; S8: The current diabetic patient makes dietary arrangements according to the dietary plan, and makes treatment arrangements according to the treatment plan, and then undergoes corresponding physical examinations after a specified period to obtain relevant examination data, and predicts the disease data at the same time point through the disease progression prediction model, and determines whether the predicted disease data and the examination data reflect that the disease corresponding to the examination data is more serious than the predicted disease. If so, the dietary plan and treatment plan are adjusted according to the specified rules, if not, no adjustment is made.

[0007] Preferably, the specific process of preprocessing the text data and image data in step S1 is as follows: S11: removing noise data from the text data and identifying abnormal data, making the text data conform to normal distribution, learning parameters of normal distribution from the text data, identifying data at low-probability points as abnormal points, and converting the abnormal points based on their previous and subsequent information; S12: removing image data whose image resolution is lower than a specified threshold value from the image data; S13: grayscale conversion is performed on the images in the remaining image data, the values ​​of the three color components of red R, green G and blue B in each image are obtained, and the average value of the three color components of red R, green G and blue B in each image frame is calculated as the grayscale conversion value; S14: performing geometric transformation processing on each image frame after the grayscale transformation processing, performing image scaling processing on each image after the geometric transformation processing, and adjusting each image to a specified size; S15: Perform image enhancement processing by adjusting image contrast and brightness, and preliminarily identify the lesion area in the image after image enhancement processing.

[0008] Preferably, the specific process of extracting text features from the text data set in step S3 is as follows: S31: extracting a single sentence from the text data set, and performing word segmentation on each single sentence, obtaining a number of words from the text data, and performing word segmentation and part-of-speech tagging on the several words, filtering out stop words therein, and generating candidate keywords; S32: constructing a candidate keyword graph based on the candidate keywords, wherein the candidate keyword graph includes a keyword node set, and constructing an edge between any two keyword nodes using a co-occurrence relationship, wherein an edge between two keyword nodes exists only when the corresponding words co-occur in a window of a specified length; S33: Based on the principle of measuring importance of the page ranking algorithm, the weights of the keyword nodes in the candidate keyword graph are initialized, and the weights of the keyword nodes are iteratively calculated until a preset number of iterations is reached or a convergence condition is met; S34: sorting the keyword node weights in reverse order, obtaining the most important designated keywords, marking the designated keywords in the text data set, and combining them into multi-word keywords when adjacent phrases are formed.

[0009] Preferably, the specific process of extracting image features from the image data at a specified time point in step S3 is as follows: S31: Gaussian blur is performed on the image after preliminary recognition, and the image is smoothed; S32: Calculate the horizontal and vertical gradient values ​​of the image after Pinghu processing. The specific formula is as follows: gx = I ( x -1, y -1) * (-1) + I ( x +1, y -1) * 1 + I (x- 1, y) * (-2) + I ( x+1, y) *2 + I ( x -1, y +1) * (-1) + I ( x +1, y +1) * 1; gy =I ( x -1, y -1) * (-1) + I ( x, y -1) * (-2) + I (x +1, y -1) * (-1) + I ( x -1, y +1) * 1 + I ( x , y +1) * 2 + I ( x +1, y +1) * 1; in, gx is the gradient value of the image in the horizontal direction, gy is the gradient value in the vertical direction of the image, I is the pixel value of the image, ( x , y ) is the image I Pixels of S33: traverse the discrete pixel points on the image, calculate a number of gradient values ​​in the horizontal direction of the image and the gradient values ​​in the vertical direction of the image to determine the gradient direction, and obtain the direction in which the pixel value changes fastest; S34: Calculate edge features of the lesion area in the image based on the obtained gradient value and the direction in which the pixel value changes fastest.

[0010] Preferably, the specific process of converting the image features into text features in step S3 is: converting into text description based on the extracted image features.

[0011] Preferably, the specific process of performing feature fusion processing on the text features based on the text data and the text features based on the image data at the same time point is as follows: S41: converting text features based on text data at a first specified time point into a first text feature vector; S42: Converting the text features based on the image data at the first specified time point to obtain a second text feature vector; S43: performing vector concatenation of the first text feature vector and the second text feature vector to obtain a fused text feature vector at the first time point; S44: Iterate steps S41 to S43 to obtain the combined text feature vectors at other specified time points.

[0012] Preferably, the specific process of selecting discrete time points between designated time points in step S5 is as follows: S51: Obtain the severity of the disease at each designated time point, and calculate the disease change rate in the time period between adjacent designated time points; S52: Allocate discrete time points of a first preset density to time periods with a disease condition change rate lower than a preset disease condition change rate, and allocate discrete time points of a second preset density to time periods with a disease condition change rate higher than a preset disease condition change rate, wherein the first preset density is smaller than the second preset density.

[0013] In a second aspect, an intelligent management system for diabetic patients based on artificial intelligence is provided, which is used to implement any one of the intelligent management methods for diabetic patients based on artificial intelligence, including a data acquisition module, a data preprocessing module, a data sorting module, a feature extraction module, a feature fusion module, a model creation module, a disease progression prediction model, a diet planning model and a diagnosis and treatment planning module; The data acquisition module is connected to the data preprocessing module, the data preprocessing module is connected to the data sorting module, the data sorting module is connected to the feature extraction module, the feature extraction module is connected to the feature fusion module, the model creation module is respectively connected to the disease progression prediction model and the diet planning model, and the disease progression prediction model is respectively connected to the diet planning model and the diagnosis and treatment planning module; The data acquisition module is used to obtain the personal information of the current diabetic patient and various examination report data at various time points; The data preprocessing module is used to classify the various inspection report data into text data and image data, and preprocess the text data and image data; The data sorting module is used to sort the preprocessed text data according to the chronological time points to construct a text data set at the specified time points; and to sort the preprocessed image data according to the chronological time points to construct an image data set corresponding to the specified time points; The feature extraction module is used to extract text features from a text data set at a specified time point, extract image features from image data at a specified time point, and perform feature conversion on the extracted image feature data to obtain text features based on the image data; The feature fusion module is used to perform feature fusion processing on the text features based on the text data and the text features based on the image data at the same time point according to the specified time point, so as to obtain the feature fusion data at each specified time point; The model creation module is used to create a disease progression prediction model and a diet planning model; The disease progression prediction model is used to predict the disease progression at discrete time points between the specified time points based on the input data of the feature fusion data at each specified time point, and to obtain the disease progression of the current diabetic patient from the first diagnosis of diabetes to the current time point based on the prediction results of the disease progression at discrete time points between the specified time points, as well as the disease progression of each time period obtained based on adjacent specified time points; The diet planning model is used to plan the diet for a specified time period starting from the current time point based on the disease progression in each time period; The diagnosis and treatment planning module is used to make diagnosis and treatment plans for a time period of a specified length starting from the current time point based on the disease progression in each time period.

[0014] The beneficial effects of the present invention include: The intelligent management method and system for diabetic patients based on artificial intelligence provided by the present invention constructs a text data set and an image data set at a specified time point; performs text feature extraction and image feature extraction, and performs feature conversion on the image feature data to obtain corresponding text features; performs feature fusion processing on the text features based on the text data and the text features based on the image data at the same time point according to the specified time point; creates and trains a disease progression prediction model, inputs feature fusion data to predict the disease at discrete time points; obtains the disease progression of the entire time period and the disease progression of each time period and performs planning and diagnosis and treatment planning; obtains inspection data after the specified period, and the disease progression prediction model predicts the disease data at the same time point, and if the disease is more serious than the predicted one, adjusts the diet and diagnosis and treatment planning. The above process realizes personalized and intelligent management of diabetic patients throughout their life cycle.

[0015] Firstly, the image data with image resolution lower than the specified threshold is removed by removing the noise data in the text data, identifying and converting the abnormal data; the images in the remaining image data are subjected to grayscale conversion; the image frames after the grayscale conversion are subjected to geometric transformation, and the images subjected to geometric transformation are subjected to image scaling to adjust each image to a specified size; the image enhancement is performed by adjusting the image contrast and brightness, and the lesion area in the image after the image enhancement is preliminarily identified, so as to provide a data basis for subsequent model training and improve the performance of the model and the accuracy of the output results.

[0016] Secondly, by extracting single sentences from the text data set and performing word segmentation on each single sentence, a number of words from the text data are obtained, and word segmentation and part-of-speech tagging are performed on the words, and stop words are filtered out to generate candidate keywords; a candidate keyword graph is constructed based on the candidate keywords, and an edge between any two keyword nodes is constructed using a co-occurrence relationship. The edge between two keyword nodes only co-occurs when their corresponding words co-occur in a window of a specified length; the weights of the keyword nodes in the candidate keyword graph are initialized; the keyword node weights are sorted in reverse order, and the most important specified keywords are obtained, and the process of combining them into multi-word keywords realizes accurate text feature extraction, which further enables subsequent models to have better performance and high accuracy of prediction results.

[0017] Thirdly, by performing Gaussian blur on the image after preliminary recognition and smoothing the image, the horizontal and vertical gradient values ​​of the image after smoothing are calculated, the discrete pixel points on the image are traversed, and a number of horizontal and vertical gradient values ​​of the image are calculated to obtain the direction in which the pixel value changes fastest. The quantitative image processing process obtains the edge features of the lesion area, which improves the accuracy of image feature extraction and provides an accurate data basis for patient condition analysis.

[0018] Again, by converting the text features based on text data at the first specified time point to obtain a first text feature vector, and converting the text features based on image data at the first specified time point to obtain a second text feature vector; the first text feature vector and the second text feature vector are vector-joined to obtain a fused text feature vector at the first time point, thereby realizing the corresponding feature fusion of text data and image data, and further providing a strong data foundation for subsequent model training and disease prediction.

[0019] Finally, the severity of the disease at each specified time point is obtained, and the rate of change of the disease in the time period between adjacent specified time points is calculated; discrete time points with high density are allocated to the time period with high disease change rate, and discrete time points with low density are allocated to the time period with low disease change rate. The discrete time point selection process between the specified time points achieves a more detailed understanding of the changes in the disease of diabetic patients in each disease time period, and also reduces the amount of calculation of the disease prediction time point through the selection of discrete time points, thereby reducing the computational complexity of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The figure is a flow chart of the intelligent management method for diabetic patients based on artificial intelligence of the present invention.

[0021] Figure 2The figure is a schematic diagram of the process of preprocessing the text data and image data according to the present invention.

[0022] Figure 3 The figure is a schematic diagram of the architecture of the intelligent management system for diabetic patients based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0023] The following is combined with Figure 1~Figure 3 The present invention is further described in detail: Example 1 See attached Figure 1 As shown, an intelligent management method for diabetic patients based on artificial intelligence includes the following steps: S1: Obtain the personal information of the current diabetic patient and various examination report data at various time points, and classify the various examination report data into text data and image data. Once a person suffers from diabetes, not only will there be glucose beyond the normal range in their urine, but over time, other organs of the body will also be damaged. Therefore, it is necessary for diabetic patients to examine other tissues of the body. Therefore, the patient's examination will have examination reports for different tissues or blood, including text data and image data.

[0024] The purpose of classifying into text data and image data is to perform feature extraction for different data types in the future, thereby increasing the efficiency and accuracy of feature extraction. The text data and image data are preprocessed to remove redundant information in the data, making the data more concise and effective, and providing an accurate data basis for subsequent model training.

[0025] S2: Sort the preprocessed text data according to the time points to construct a text data set at the specified time points; sort the preprocessed image data according to the time points to construct an image data set corresponding to the specified time points. The purpose of sorting the text data and image data according to the time points is to obtain the disease progression according to the time sequence, to facilitate the acquisition of the disease changes of diabetic patients in each time period, and to provide a certain data basis for subsequent diet planning and diagnosis and treatment planning.

[0026] S3: Perform text feature extraction on the text data set at the specified time point, perform image feature extraction on the image data at the specified time point, perform feature conversion on the extracted image feature data to obtain text features based on the image data, and convert the image features into text features to convert the extracted image features into text descriptions.

[0027] S4: Perform feature fusion processing on the text features based on text data and the text features based on image data at the same time point at the specified time point to obtain the feature fusion data at each specified time point. The performance and generalization ability of the model are improved through the process of feature fusion. By merging the features of text features converted from multiple different source image features and text features extracted from text data into a better fused feature representation, feature fusion can significantly improve the classification accuracy, robustness and generalization ability during model training, while reducing the risk of overfitting. Because different features can capture different information. When some features fail, other features can make up for their shortcomings, thereby improving the robustness of the model.

[0028] S5: Create a disease progression prediction model and train it with the labeled data of diabetic patients, input the feature fusion data of each designated time point into the disease progression prediction model to predict the disease at discrete time points between the designated time points. By expanding the discrete time points and combining the disease prediction of the disease progression prediction model, it is convenient to obtain a more detailed disease progression status of the disease, providing a data basis for subsequent diet planning and diagnosis and treatment planning.

[0029] S6: Based on the disease prediction results of discrete time points between the specified time points, the disease progression of the current diabetic patient from the first diagnosis of diabetes to the current time point is obtained, as well as the disease progression of each time period obtained based on adjacent specified time points.

[0030] S7: Create a diet planning model and train it based on the labeled diet planning data, and plan the diet for a specified time period starting from the current time point based on the disease progression in each time period, and plan the diagnosis and treatment for a specified time period starting from the current time point based on the disease progression in each time period; S8: The current diabetic patient makes dietary arrangements according to the dietary plan, and makes treatment arrangements according to the treatment plan, and then undergoes corresponding physical examinations after a specified period to obtain relevant examination data, and predicts the disease data at the same time point through the disease progression prediction model, and determines whether the predicted disease data and the examination data reflect that the disease corresponding to the examination data is more serious than the predicted disease. If so, the dietary plan and treatment plan are adjusted according to the specified rules, if not, no adjustment is made.

[0031] In this embodiment, by constructing a text data set and an image data set at a specified time point; performing text feature extraction and image feature extraction, and performing feature conversion on the image feature data to obtain corresponding text features; performing feature fusion processing on the text features based on the text data and the text features based on the image data at the same time point at the specified time point; creating a disease progression prediction model and training it, inputting feature fusion data to predict the disease at discrete time points; obtaining the disease progression of the entire time period and the disease progression of each time period and performing planning and diagnosis and treatment planning; obtaining inspection data after the specified period, and the disease progression prediction model predicts the disease data at the same time point, and if the disease is more serious than the predicted one, adjusting the diet and diagnosis and treatment plan. The above process realizes personalized and intelligent management of diabetic patients throughout their life cycle.

[0032] Example 2 Based on Example 1, the specific process of preprocessing the text data and image data in step S1 is as follows: S11: removing noise data from the text data and identifying abnormal data, making the text data conform to normal distribution, learning parameters of normal distribution from the text data, identifying data at low-probability points as abnormal points, and converting the abnormal points based on their previous and subsequent information; S12: removing image data whose image resolution is lower than a specified threshold value from the image data; S13: grayscale conversion is performed on the images in the remaining image data, the values ​​of the three color components of red R, green G and blue B in each image are obtained, and the average value of the three color components of red R, green G and blue B in each image frame is calculated as the grayscale conversion value; S14: performing geometric transformation processing on each image frame after the grayscale transformation processing, performing image scaling processing on each image after the geometric transformation processing, and adjusting each image to a specified size; S15: Perform image enhancement processing by adjusting image contrast and brightness, and preliminarily identify the lesion area in the image after image enhancement processing.

[0033] By removing the noise data in the text data, identifying and converting the abnormal data, the image data with image resolution lower than the specified threshold is removed; the images in the remaining image data are subjected to grayscale conversion; geometric transformation is performed on each image frame after the grayscale conversion, and image scaling is performed on each image subjected to the geometric transformation, so that each image is adjusted to a specified size; image enhancement is performed by adjusting the image contrast and brightness, and the lesion area in the image after the image enhancement is preliminarily identified, so as to provide a data basis for subsequent model training and improve the performance of the model and the accuracy of the output results.

[0034] Example 3 Based on Example 1 or Example 2, see Figure 2 As shown, the specific process of extracting text features from the text dataset in step S3 is as follows: S31: extracting a single sentence from the text data set, and performing word segmentation on each single sentence, obtaining a number of words from the text data, and performing word segmentation and part-of-speech tagging on the several words, filtering out stop words therein, and generating candidate keywords; S32: constructing a candidate keyword graph based on the candidate keywords, wherein the candidate keyword graph includes a keyword node set, and constructing an edge between any two keyword nodes using a co-occurrence relationship, wherein an edge between two keyword nodes exists only when the corresponding words co-occur in a window of a specified length; S33: Based on the principle of measuring importance of the page ranking algorithm, the weights of the keyword nodes in the candidate keyword graph are initialized, and the weights of the keyword nodes are iteratively calculated until a preset number of iterations is reached or a convergence condition is met; S34: sorting the keyword node weights in reverse order, obtaining the most important designated keywords, marking the designated keywords in the text data set, and combining them into multi-word keywords when adjacent phrases are formed.

[0035] The method extracts single sentences from the text data set, performs word segmentation on each single sentence, obtains several words from the text data, performs word segmentation and part-of-speech tagging on the several words, and generates candidate keywords after filtering out stop words; constructs a candidate keyword graph based on the candidate keywords, and uses co-occurrence relationships to construct edges between any two keyword nodes, where the edges between two keyword nodes co-occur only when their corresponding words co-occur in a window of a specified length; initializes the weights of the keyword nodes in the candidate keyword graph; sorts the keyword node weights in reverse order, obtains the most important specified keywords, and combines them into a multi-word keyword process, thereby achieving accurate text feature extraction, and further enabling subsequent models to have better performance and high accuracy of prediction results.

[0036] In this embodiment, the specific process of extracting image features from the image data at a specified time point in step S3 is as follows: S31: Gaussian blur is performed on the image after preliminary recognition, and the image is smoothed; S32: Calculate the horizontal and vertical gradient values ​​of the image after Pinghu processing. The specific formula is as follows: gx = I ( x -1, y -1) * (-1) +I ( x +1, y -1) * 1 + I (x- 1, y) * (-2) + I ( x +1, y) *2 + I ( x -1, y +1) * (-1) + I ( x +1, y +1) * 1; gy =I ( x -1, y -1) * (-1) + I ( x, y -1) * (-2) + I (x +1, y -1) * (-1) + I ( x -1, y +1) * 1 + I ( x , y +1) * 2 + I ( x +1, y +1) * 1; in, gx is the gradient value of the image in the horizontal direction, gy is the gradient value in the vertical direction of the image, I is the pixel value of the image, ( x , y ) is the image I Pixels of S33: traverse the discrete pixel points on the image, calculate a number of gradient values ​​in the horizontal direction of the image and the gradient values ​​in the vertical direction of the image to determine the gradient direction, and obtain the direction in which the pixel value changes fastest; S34: Calculate edge features of the lesion area in the image based on the obtained gradient value and the direction in which the pixel value changes fastest.

[0037] In this embodiment, by performing Gaussian blur on the image after preliminary recognition and smoothing the image, calculating the gradient values ​​of the image in the horizontal and vertical directions after the smoothing process, traversing the discrete pixel points on the image, and calculating a number of gradient values ​​in the horizontal direction and the vertical direction of the image, the direction in which the pixel value changes fastest is obtained, and the edge features of the lesion area are obtained through a quantization-based image processing process, which improves the accuracy of image feature extraction and provides an accurate data basis for the analysis of the patient's condition.

[0038] Example 4 On the basis of Example 1 or Example 2 or Example 3, the specific process of performing feature fusion processing on the text features based on the text data and the text features based on the image data at the same time point is as follows: S41: Convert the text features based on the text data at the first specified time point into a first text feature vector. The text feature vector obtained by conversion is convenient for subsequent computer or model recognition and understanding; converting the text features according to the time point is to facilitate subsequent stage-by-stage disease analysis.

[0039] S42: Converting the text features based on the image data at the first specified time point to obtain a second text feature vector; S43: performing vector concatenation of the first text feature vector and the second text feature vector to obtain a fused text feature vector at the first time point; S44: Iterate steps S41 to S43 to obtain the combined text feature vectors at other specified time points.

[0040] In this embodiment, the first text feature vector is obtained by converting the text features based on text data at the first specified time point, and the second text feature vector is obtained by converting the text features based on image data at the first specified time point; the first text feature vector and the second text feature vector are vector-joined to obtain the fused text feature vector at the first time point, thereby realizing the corresponding feature fusion of text data and image data, and further providing a strong data foundation for subsequent model training and disease prediction.

[0041] The specific process of selecting discrete time points between designated time points in step S5 is as follows: S51: Obtain the severity of the disease at each designated time point, calculate the disease change rate in the time period between adjacent designated time points, and obtain the disease change situation in unit time; S52: Allocate discrete time points of a first preset density to time periods with a disease condition change rate lower than a preset disease condition change rate, and allocate discrete time points of a second preset density to time periods with a disease condition change rate higher than a preset disease condition change rate, wherein the first preset density is smaller than the second preset density.

[0042] In this embodiment, the severity of the disease at each designated time point is obtained, and the rate of change of the disease in the time period between adjacent designated time points is calculated; discrete time points with high density are allocated to the time period with high disease change rate, and discrete time points with low density are allocated to the time period with low disease change rate. The discrete time point selection process between the designated time points achieves a more detailed understanding of the changes in the disease of diabetic patients in each disease time period, and also reduces the amount of calculation of the disease prediction time point through the selection of discrete time points, thereby reducing the calculation complexity of the model.

[0043] An intelligent management system for diabetic patients based on artificial intelligence, used to implement any one of the intelligent management methods for diabetic patients based on artificial intelligence, see Figure 3 As shown, it includes a data acquisition module, a data preprocessing module, a data sorting module, a feature extraction module, a feature fusion module, a model creation module, a disease progression prediction model, a diet planning model and a diagnosis and treatment planning module.

[0044] The data acquisition module is connected to the data preprocessing module, the data preprocessing module is connected to the data sorting module, the data sorting module is connected to the feature extraction module, the feature extraction module is connected to the feature fusion module, the model creation module is respectively connected to the disease progression prediction model and the diet planning model, and the disease progression prediction model is respectively connected to the diet planning model and the diagnosis and treatment planning module.

[0045] The data acquisition module is used to obtain the personal information of the current diabetic patient and the data of various examination reports at various time points; the data preprocessing module is used to classify the data of various examination reports into text data and image data, and preprocess the text data and image data; the data sorting module is used to sort the preprocessed text data according to the chronological time points to construct a text data set at the specified time point; sort the preprocessed image data according to the chronological time points to construct an image data set corresponding to the specified time point; the feature extraction module is used to extract text features from the text data set at the specified time point, extract image features from the image data at the specified time point, and perform feature conversion on the extracted image feature data to obtain text features based on the image data.

[0046] A feature fusion module is used to perform feature fusion processing on the text features based on text data and the text features based on image data at the same time point according to the specified time point, so as to obtain the feature fusion data of each specified time point; a model creation module is used to create a disease progression prediction model and a diet planning model; the disease progression prediction model is used to predict the disease at discrete time points between the specified time points based on the input data of the feature fusion data of each specified time point, and obtain the disease progression of the current diabetic patient from the first diagnosis of diabetes to the current time point based on the prediction results of the disease at discrete time points between the specified time points, as well as the disease progression of each time period obtained based on adjacent specified time points; a diet planning model is used to plan the diet for a specified length time period starting from the current time point based on the disease progression of each time period; a diagnosis and treatment planning module is used to make a diagnosis and treatment plan for a specified length time period starting from the current time point based on the disease progression of each time period.

[0047] In summary, the artificial intelligence-based intelligent management method and system for diabetic patients provided by the present invention removes noise data in text data, identifies abnormal data, and performs conversion processing to remove image data with image resolution lower than a specified threshold in image data; performs grayscale conversion processing on images in the remaining image data; performs geometric transformation processing on each image frame after the grayscale conversion processing, performs image scaling processing on each image after the geometric transformation processing, and adjusts each image to a specified size; performs image enhancement processing by adjusting image contrast and brightness, and preliminarily identifies the lesion area in the image after the image enhancement processing, thereby providing a data basis for subsequent model training and improving the performance of the model and the accuracy of the output results.

[0048] By extracting a single sentence from the text data set, and performing word segmentation on each single sentence, obtaining several words of the text data, and performing word segmentation and part-of-speech tagging on several words, and filtering out the stop words, candidate keywords are generated; a candidate keyword graph is constructed based on the candidate keywords, and the co-occurrence relationship is used to construct the edge between any two keyword nodes, and the edge between the two keyword nodes only co-occurs in the window of the specified length corresponding to the words; the keyword node weights in the candidate keyword graph are initialized; the keyword node weights are sorted in reverse order, the most important specified keywords are obtained, and the process of combining into multi-word keywords realizes accurate text feature extraction, and further provides the subsequent model with better performance and high accuracy of prediction results. By performing Gaussian blur on the image after preliminary recognition, and smoothing the image, calculating the horizontal and vertical gradient values ​​of the image after Pinghu processing, traversing the discrete pixel points on the image, calculating several horizontal gradient values ​​of the image and the vertical gradient values ​​of the image, obtaining the direction with the fastest pixel value change, and obtaining the edge features of the lesion area based on quantization image processing process, the accuracy of image feature extraction is improved, and an accurate data basis is provided for the patient's condition analysis.

[0049] The first text feature vector is obtained by converting the text feature based on text data at the first specified time point, and the second text feature vector is obtained by converting the text feature based on image data at the first specified time point; the first text feature vector and the second text feature vector are vector-joined to obtain the fused text feature vector of the first time point, thereby realizing the corresponding feature fusion of text data and image data, and further providing a strong data basis for subsequent model training and disease prediction. The severity of the disease at each specified time point is obtained, and the disease change rate in the time period between adjacent specified time points is calculated; the discrete time point selection process between the specified time points is performed by allocating high-density discrete time points to the time period with high disease change rate, and low-density discrete time points to the time period with low disease change rate, thereby realizing a more detailed understanding of the disease change of the diabetic patient in each disease time period, and also reducing the amount of calculation of the disease prediction time point and the calculation complexity of the model through the selection of discrete time points.

Claims

1. An intelligent management method for diabetic patients based on artificial intelligence, characterized in that: The following steps are involved: S1: Obtaining personal information of the current diabetic patient and various examination report data at various time points, classifying the various examination report data into text data and image data, and preprocessing the text data and image data; S2: sorting the preprocessed text data according to the time points, and constructing a text data set at the specified time point; sorting the preprocessed image data according to the time points, and constructing an image data set corresponding to the specified time point; S3: extracting text features from the text data set at a specified time point, extracting image features from the image data at a specified time point, and performing feature conversion on the extracted image feature data to obtain text features based on the image data; S4: performing feature fusion processing on the text features based on the text data and the text features based on the image data at the same time point at the specified time point to obtain feature fusion data at each specified time point; S5: creating a disease progression prediction model and training it with the labeled data of diabetic patients, inputting the feature fusion data of each designated time point into the disease progression prediction model to predict the disease at discrete time points between the designated time points; S6: based on the disease prediction results of discrete time points between the specified time points, obtaining the disease progression of the current diabetic patient from the first diagnosis of diabetes to the current time point, and the disease progression of each time period obtained based on adjacent specified time points; S7: Create a diet planning model and train it based on the labeled diet planning data, and plan the diet for a specified time period starting from the current time point based on the disease progression in each time period, and plan the diagnosis and treatment for a specified time period starting from the current time point based on the disease progression in each time period; S8: The current diabetic patient makes dietary arrangements according to the dietary plan, and makes treatment arrangements according to the treatment plan, and then undergoes corresponding physical examinations after a specified period to obtain relevant examination data, and predicts the disease data at the same time point through the disease progression prediction model, and determines whether the predicted disease data and the examination data reflect that the disease corresponding to the examination data is more serious than the predicted disease. If so, the dietary plan and treatment plan are adjusted according to the specified rules, if not, no adjustment is made.

2. The intelligent management method for diabetic patients based on artificial intelligence according to claim 1, characterized in that: The specific process of preprocessing the text data and image data in step S1 is as follows: S11: removing noise data from the text data and identifying abnormal data, making the text data conform to normal distribution, learning parameters of normal distribution from the text data, identifying data at low-probability points as abnormal points, and converting the abnormal points based on their previous and subsequent information; S12: removing image data whose image resolution is lower than a specified threshold value from the image data; S13: grayscale conversion is performed on the images in the remaining image data, the values ​​of the three color components of red R, green G and blue B in each image are obtained, and the average value of the three color components of red R, green G and blue B in each image frame is calculated as the grayscale conversion value; S14: performing geometric transformation processing on each image frame after the grayscale transformation processing, performing image scaling processing on each image after the geometric transformation processing, and adjusting each image to a specified size; S15: Perform image enhancement processing by adjusting image contrast and brightness, and preliminarily identify the lesion area in the image after image enhancement processing.

3. The intelligent management method for diabetic patients based on artificial intelligence according to claim 2, characterized in that: The specific process of extracting text features from the text dataset in step S3 is as follows: S31: extracting a single sentence from the text data set, and performing word segmentation on each single sentence, obtaining a number of words from the text data, and performing word segmentation and part-of-speech tagging on the several words, filtering out stop words therein, and generating candidate keywords; S32: constructing a candidate keyword graph based on the candidate keywords, wherein the candidate keyword graph includes a keyword node set, and constructing an edge between any two keyword nodes using a co-occurrence relationship, wherein an edge between two keyword nodes exists only when the corresponding words co-occur in a window of a specified length; S33: Based on the principle of measuring importance of the page ranking algorithm, the weights of the keyword nodes in the candidate keyword graph are initialized, and the weights of the keyword nodes are iteratively calculated until a preset number of iterations is reached or a convergence condition is met; S34: sorting the keyword node weights in reverse order, obtaining the most important designated keywords, marking the designated keywords in the text data set, and combining them into multi-word keywords when adjacent phrases are formed.

4. The intelligent management method for diabetic patients based on artificial intelligence according to claim 2, characterized in that: The specific process of extracting image features from the image data at a specified time point in step S3 is as follows: S31: Gaussian blur is performed on the image after preliminary recognition, and the image is smoothed; S32: Calculate the horizontal and vertical gradient values ​​of the image after Pinghu processing. The specific formula is as follows: gx = I ( x -1, y -1) * (-1) + I ( x +1, y -1) * 1 + I (x- 1, and) * (-2) + I ( x +1,y) * 2 + I ( x -1, y +1) * (-1) + I ( x +1, y +1) * 1; gy =I ( x -1, y -1) * (-1) + I ( x, y -1) * (-2) + I (x +1, y -1) * (-1) + I ( x -1, y +1) * 1 + I ( x , y +1) * 2 + I ( x +1, y +1) * 1； in, gx is the horizontal gradient value of the image, gy is the gradient value in the vertical direction of the image, I is the pixel value of the image, ( x , y ) is the image I Pixels of S33: traverse the discrete pixel points on the image, calculate a number of gradient values ​​in the horizontal direction of the image and the gradient values ​​in the vertical direction of the image to determine the gradient direction, and obtain the direction in which the pixel value changes fastest; S34: Calculate edge features of the lesion area in the image based on the obtained gradient value and the direction in which the pixel value changes fastest.

5. The intelligent management method for diabetic patients based on artificial intelligence according to claim 4, characterized in that: The specific process of converting the image features into text features in step S3 is: converting the extracted image features into text descriptions.

6. The intelligent management method for diabetic patients based on artificial intelligence according to claim 1, characterized in that: The specific process of fusing the text features based on text data with the text features based on image data at the same time point is as follows: S41: converting text features based on text data at a first specified time point into a first text feature vector; S42: Converting the text features based on the image data at the first specified time point to obtain a second text feature vector; S43: performing vector concatenation of the first text feature vector and the second text feature vector to obtain a fused text feature vector at the first time point; S44: Iterate steps S41 to S43 to obtain the combined text feature vectors at other specified time points.

7. The intelligent management method for diabetic patients based on artificial intelligence according to claim 1, characterized in that: The specific process of selecting discrete time points between designated time points in step S5 is as follows: S51: Obtain the severity of the disease at each designated time point, and calculate the disease change rate in the time period between adjacent designated time points; S52: Allocate discrete time points of a first preset density to time periods with a disease condition change rate lower than a preset disease condition change rate, and allocate discrete time points of a second preset density to time periods with a disease condition change rate higher than a preset disease condition change rate, wherein the first preset density is smaller than the second preset density.

8. An intelligent management system for diabetic patients based on artificial intelligence, used to implement an intelligent management method for diabetic patients based on artificial intelligence according to any one of claims 1 to 7, characterized in that: It includes data acquisition module, data preprocessing module, data sorting module, feature extraction module, feature fusion module, model creation module, disease progression prediction model, diet planning model and diagnosis and treatment planning module; The data acquisition module is connected to the data preprocessing module, the data preprocessing module is connected to the data sorting module, the data sorting module is connected to the feature extraction module, the feature extraction module is connected to the feature fusion module, the model creation module is respectively connected to the disease progression prediction model and the diet planning model, and the disease progression prediction model is respectively connected to the diet planning model and the diagnosis and treatment planning module; The data acquisition module is used to obtain the personal information of the current diabetic patient and various examination report data at various time points; The data preprocessing module is used to classify the various inspection report data into text data and image data, and preprocess the text data and image data; The data sorting module is used to sort the preprocessed text data according to the chronological time points to construct a text data set at the specified time points; and to sort the preprocessed image data according to the chronological time points to construct an image data set corresponding to the specified time points; The feature extraction module is used to extract text features from a text data set at a specified time point, extract image features from image data at a specified time point, and perform feature conversion on the extracted image feature data to obtain text features based on the image data; The feature fusion module is used to perform feature fusion processing on the text features based on the text data and the text features based on the image data at the same time point according to the specified time point, so as to obtain the feature fusion data at each specified time point; The model creation module is used to create a disease progression prediction model and a diet planning model; The disease progression prediction model is used to predict the disease progression at discrete time points between the specified time points based on the input data of the feature fusion data at each specified time point, and to obtain the disease progression of the current diabetic patient from the first diagnosis of diabetes to the current time point based on the prediction results of the disease progression at discrete time points between the specified time points, as well as the disease progression of each time period obtained based on adjacent specified time points; The diet planning model is used to plan the diet for a specified time period starting from the current time point based on the disease progression in each time period; The diagnosis and treatment planning module is used to make diagnosis and treatment plans for a time period of a specified length starting from the current time point based on the disease progression in each time period.

Images