Adaptive generation system of diabetes diet plan based on deep learning

Abstract

The present invention discloses a self-adaptive generation system of a diabetes diet plan based on deep learning, including: a collection and storage module, configured to collect and produce data on the dynamic changes of the total calories of patients, and construct a personalized data set for generating a calorie supplement plan according to the basal metabolic rate, occupational consumption, exercise consumption, dietary supplement, and the difference between the waist-to-hip ratio and the mean of fasting blood glucose; a calculation module, which imports the personalized data set into an improved linear regression model and performs data training to obtain the corresponding optimal weight of each data in the data set; an evaluation and feedback module, which automatically calculates the dietary calorie supplement value according to the optimal weight and deviation value of the calculation module by setting up a dual gradient target, generates a personalized diet plan, and timely feedback optimization. The present invention constructs a personalized data set of the total calorie changes of patients, and introduces a deep learning method into the formulation of the patient's diet plan, providing more accurate and scientific dietary guidance for diabetic patients.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of medical health technology, and in particular to a deep learning-based adaptive generation system for a diabetes diet plan. Background Art

[0002] Diabetes is a chronic metabolic disease with a high incidence rate. The quality of intervention in the patient's lifestyle (diet, exercise, etc.) directly affects the progression of the disease and the effectiveness of rehabilitation. The traditional saying of "control your mouth and move your legs" is a vivid expression. At present, the dietary management of diabetic patients is mainly based on daily calorie consumption. Specifically, it is to simply estimate the calorie consumption based on the comparison of the patient's weight with the standard weight, physical activity and other indicators, and propose a fuzzy and uniform diet plan. In fact, this management method is very extensive. It lacks precise dietary guidance for individual changes in the patient's condition, and lacks dynamic calorie consumption assessment and corresponding information-based automatic adjustment technology.

[0003] The purpose of daily dietary management for patients is to alleviate the condition of diabetes and reduce the damage of related complications. This complicated and tedious rehabilitation work depends on precise dietary guidance. In recent years, clinical research on diabetes has emphasized the targeted significance of waist-to-hip ratio and blood sugar in the entire course of disease management, but this has not been fully reflected in dietary rehabilitation. The current popular dietary guidance program not only ignores the improvement of the patient's own physiological characteristics (such as visceral fat content, muscle content and other indicators), but also lacks a direct connection with the control target of blood sugar levels. At the same time, it also lacks accurate evaluation, feedback and adjustment methods that can reflect the effect of dietary intervention, resulting in the improvement of the dietary rehabilitation effect of diabetic patients. An important reason for this is the lack of innovation in effective information technology processing methods that can achieve rapid response to specific patients.

[0004] At present, the common practice of formulating diabetes dietary guidance programs is to follow the following steps, as follows:

[0005] (1) Simple estimation of ideal body weight: standard body weight (kg) = height (cm) - 105. If the standard body weight is 20% lower than the standard body weight, it is considered to be thin; if the standard body weight is 20% higher than the standard body weight, it is considered to be obese; if the standard body weight is ±10%, it is considered to be ideal body weight.

[0006] (2) Calculate the total calories needed per day: Calculate the total calories needed from food per day based on the standard body weight and physical labor. The calories needed per day = standard body weight × calories needed per kilogram of body weight. For example, if a person is 167 cm tall, weighs 75 kg, and does moderate physical labor, the standard body weight is 167-105 = 62 (kg). If the actual body weight exceeds the standard body weight by 20%, it is considered obese; the calories needed per day are 62 × 30 = 1860 kcal (Table 1).

[0007] body shape Bed rest Light physical labor Medium physical labor Heavy physical labor thin 20～25 35 40 40～45 normal 15～20 30 35 40 obesity 15 20～25 30 35

[0008] Table 1: Calories required per kilogram of body weight for different physical labors [kcal / body weight (kg)]

[0009] (3) Select a recipe with the corresponding calories: All foods are raw before cooking (Table 2)

[0010]

[0011]

[0012] Table 2: Calorie-to-calorie diet

[0013] For example: male, 56 years old, 170cm, 85kg, accountant. Ideal weight is 170-105=65kg; actual weight is 85kg, 30% higher than standard weight, which is considered obese. Accounting is a light physical labor. According to Table 1, the daily energy intake standard is 20-25kcal / kg, so the total calories required for the whole day are 65×20-25=1300-1625kcal. Therefore, a 1400kcal diet is recommended.

[0014] Food exchange table: Only foods of the same type can be exchanged, and the weight of all foods is the weight of 1 exchange portion (90kcal). (Table 3)

[0015]

[0016]

[0017]

[0018] Table 3: Food Exchange Table

[0019] However, this method of formulating a diet plan has obvious defects: first, the degree of personalization is low. Calorie consumption is estimated based on the general type of physical activity and body weight, and the average calorie demand of the population is referred to for increase or decrease. Only the patient's daily physical activity is taken into account, but the patient's additional physical exercise, blood sugar fluctuations, etc. are not taken into account. It is difficult to comprehensively evaluate calorie consumption and it is difficult to accurately supplement calories according to the actual needs of individuals to alleviate their condition. Second, it is inefficient. Each physician or nutritionist is limited by the number of patients they serve. In the case of difficulty in dynamically evaluating individual calorie consumption, it is impossible to provide accurate dietary guidance to most patients. Third, the effect is poor. Most of the current diet plan formulation plans are aimed at controlling calorie intake, while ignoring the improvement of the patient's own physiological characteristics (such as visceral fat content, muscle content and other indicators) and the control target of blood sugar levels, resulting in generally poor results in dietary rehabilitation. Fourth, there is a lack of feedback mechanism. Changes in the patient's daily life (exercise, diet, etc.) cannot be linked in a timely manner, and it is difficult to accurately implement dynamic evaluation and make dietary adjustments based on the fluctuations in the patient's daily calorie consumption.

[0020] Therefore, in response to the above-mentioned shortcomings, it is necessary to achieve personalization, precision, efficiency and automation in diabetic dietary guidance. Relying on technological innovation to greatly improve the effectiveness of patient dietary management is an urgent task facing the industry. Summary of the invention

[0021] The purpose of the present invention is to provide a diabetes diet guidance system based on deep learning, which uses waist-to-hip ratio (WHR), a sensitive indicator for reflecting the patient's visceral fat content, and fasting blood glucose mean (FPG) to set dual gradient targets, more accurately derive the patient's calorie supplement plan, and automatically generate a diet plan in a timely manner. A personalized data set of the dynamic changes in the patient's total calories is constructed, and based on this, a generation system for diabetes diet guidance plans is created using an improved linear regression model.

[0022] The embodiment of the present invention provides a system for adaptively generating a diabetes diet plan based on deep learning, characterized by comprising: a collection and storage module configured to collect and generate data on the dynamic changes in the total calories of a patient, and to construct a personalized data set for machine learning training and generating a calorie supplement plan based on the basal metabolic rate, occupational consumption, exercise consumption, dietary supplement amount, and the difference between the waist-to-hip ratio and the mean of fasting blood sugar;

[0023] A calculation module, configured as an improved linear regression model, imports the patient personalized data set into the model and performs data training by a program, thereby obtaining the optimal weight corresponding to each data in the data set;

[0024] The evaluation and feedback module is configured to automatically calculate the calorie supplement value of the optimal solution based on the optimal weights and deviation values ​​described in the calculation module by setting up cWHR gradient targets and cFPG gradient targets, so as to generate a personalized diet plan and provide timely feedback for optimization.

[0025] The optimal solution for the heat supplement plan, the linear regression model is: the output feature value is y = input feature x j and model weights w j After matrix-vector multiplication + bias value k, that is:

[0026]

[0027] Among them, the output characteristic value y is the double gradient target value. m >0.9, female WHR f When >0.85, y is the waist-to-hip ratio gradient target value cWHR. Input characteristic value x i is the value of each data in the data set (basal metabolic rate, occupational consumption, exercise consumption, dietary supplement), and the model weight w i is the weight of the ith item.

[0028] When the waist-to-hip ratio (WHR) of men m ≤0.9, female waist-to-hip ratio WHR f When ≤0.85, y is the fasting blood glucose gradient target value cFPG. Input characteristic value x j is the value of each data in the data set (basal metabolic rate, occupational consumption, exercise consumption, dietary supplement), and the model weight w j is the j-th weight.

[0029] The optimal solution for the calorie supplementation plan, the collection and storage module is configured to dynamically collect and update the relevant data of the patient's total calorie consumption, continuously form a new data set, and transmit the new data set to the calculation module and the evaluation module for processing in turn, and push the updated dietary calorie supplementation recommendation according to the instant optimal weight;

[0030] When dynamically updating the heat supplementation plan, there is a gradient of the loss function. When the gradient decreases, the weight is updated in the opposite direction of the gradient. The weights w are obtained according to the dual gradient target background. i 、w j And the optimal value of the scalar deviation value k includes:

[0031] Add the bias value to the weight: w←[w,k];

[0032] Calculate the square loss: e(x, y, w) = 1 / 2n||y-xw|| 2 ;

[0033] n means there are n samples in the matrix, the square loss e is the error between the sample prediction value and the actual value. The smaller the value, the smaller the loss. w is the matrix form of the weight.

[0034] The optimal solution for the heat supplementation plan is to set the batch size b to the length of the patient's personalized data set and set is the learning rate. After traversing the set number of times, when e is less than the set threshold, the judgment result is valid, and the data set weight w and scalar deviation value k are output. When k is greater than the set threshold, the judgment result is invalid. This function accepts the model parameter set x, y, w, and b as input, the size of each update step is given by Decide;

[0035]

[0036] The optimal solution for the calorie supplementation plan, the relevant data include BM basal metabolic rate, OM occupational calorie consumption, MM exercise calorie consumption and DP dietary calorie supplementation. Among them, basal metabolic rate BM = BMR * 0.2389, BMR is the basal metabolic rate, calculated according to age, height and weight; occupational consumption OM = h * 8 * (BM / 24), h is the energy consumption of very light activity, light activity, moderate physical activity, heavy physical activity and very heavy physical activity represented by the patient's different occupations, which are approximately 1.2, 1.5, 4, 6 and 10 times the basal metabolic rate, respectively.

[0037] The optimal solution for the heat supplementation plan, the process also includes a data standardization process, taking the base 10 logarithm of BM, OM, MM and then negating it, taking the base 10 logarithm of DP, multiplying cWHR by 100, and cFPG does not need to be changed, so that these parameter values ​​are all kept in a similar range that is convenient for calculation and processing, that is:

[0038]

[0039]

[0040]

[0041]

[0042] Wlq L =cWHR×100

[0043] The waist-to-hip ratio WHR=W / H, and cWHR is the difference in waist-to-hip ratios at different time points.

[0044] The present invention has the following effects:

[0045] (1) The present invention collects and produces data on the dynamic changes in the total calories of patients in the early stage, constructs a personalized data set that can be used for machine learning training and generate a calorie supplement plan, and then establishes an improved linear regression model related to deep learning. It can be combined with the gradient target of patient diet management to more accurately obtain the patient's calorie supplement value and automatically generate a diet plan.

[0046] (2) The present invention can combine the patient's personal physical data, occupation, and lifestyle habits such as exercise and diet, accurately grasp the patient's calorie consumption characteristics through program learning and training, and then automatically adjust the patient's diet plan through timely feedback.

[0047] (3) The present invention targets the patient's cWHR and cFPG within a period of time, and effectively links the dietary rehabilitation process to important indicators such as the patient's visceral fat, body weight, waist circumference, hip circumference, and fasting blood glucose. This is more in line with the new consensus and new trend of clinical research, and can more accurately and quickly achieve disease remission in diabetic patients.

[0048] (4) The present invention selects BM (basal metabolic rate), OM (occupational calories), MM (exercise calories), and DP (dietary calories) related to cWHR and cFPG as the main basis for modeling, and dynamically calculates the optimal weights of the four indicators through automatic training of the program. This not only more accurately reflects the actual energy consumption and dietary supplementation of patients, but is also simpler and more practical than traditional dietary calculation methods.

[0049] (5) The present invention is based on improved linear regression data processing, and the alternating collection and analysis of relevant indicators can achieve dynamic feedback and synchronize the push of dietary guidance, which greatly simplifies the cumbersome clinical dietary plan formulation process and improves the efficiency of patient dietary management.

[0050] (6) The optimal caloric supplement value calculated based on deep learning can be combined with the fluctuation of the patient's fasting blood glucose level to provide a basic basis for adjusting the patient's carbohydrate, protein and fat intake and proportion. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0052] Figure 1 This is a schematic diagram of the module structure of the deep learning-based adaptive generation system for a diabetes diet plan in Embodiment 1 of the present invention;

[0053] Figure 2 This is a schematic diagram of the operation flow of the system for adaptively generating a diabetes diet plan based on deep learning in the second embodiment of the present invention; DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0056] Example 1

[0057] like Figure 1 and Figure 2 As shown, the self-adaptive generation system of diabetes diet plan based on deep learning provided by the present invention comprises: a collection and storage module, configured to collect and produce data of dynamic changes in total calories of patients, and construct personalized data for machine learning training and generation of calorie supplement plan according to basal metabolic rate, occupational consumption, exercise consumption, dietary supplement amount, and difference between waist-to-hip ratio and fasting blood sugar mean;

[0058] As a specific embodiment of the present invention, patient data is collected within a time period, including data such as gender, age, height, weight, waist circumference, hip circumference, fasting blood sugar, work intensity, exercise intensity and diet. Among them, waist circumference and hip circumference are collected to calculate WHR (waist-to-hip ratio, waist / hip circumference), height, weight and age are collected to calculate BM, work intensity is collected to calculate OM, exercise intensity data is collected to calculate MM, diet data is collected to calculate DP, and fasting blood sugar is collected to calculate FPG. Specifically: when the patient registers on the client, he fills in his name, gender, age, height, weight, waist circumference, hip circumference, fasting blood sugar, work intensity and exercise intensity data on the client, and the client generates the first diet plan. During continued use, weight, waist circumference, fasting blood sugar and hip circumference are regularly updated, and the actual exercise execution amount and the actual diet plan execution amount are fed back daily, which are sent to the calculation module by the client according to the data requirements in the table below.

[0059]

[0060]

[0061] Table 4: Patient data table calculation module, configured as a linear regression model based on a single-layer neural network, imports the personalized data set into the model, and obtains the optimal weights of each data in the data set based on program learning and training. The calculation module uses SQL statements to query all historical data of the patient in the database according to the patient ID, including gender, age, height, weight, waist circumference, hip circumference, fasting blood sugar, work intensity, exercise program (if any) and diet program, calculates the patient's WHR value, and sorts it in the order of measurement date. Traverse all WHR values ​​and make a difference. The difference defines cWHR. When the difference is greater than 0.01, define the time period as a single data set, and calculate BM, OM, MM, DP and cWHR within the time period, and store the data set at the same time for direct call next time. Similarly, traverse all FPG means and make a difference. The difference defines cFPG. When the difference is greater than 0.1, define the time period as a single data set, and calculate BM, OM, MM, DP and cFPG within the time period, and store the data set at the same time for direct call next time.

[0062] Calculate waist-to-hip ratio WHR = W / H

[0063] W waist unit: cm; H hip unit: cm

[0064] Calculate the waist-to-hip ratio difference cWHR, the difference in WHR over a period of time, for example, the WHR on October 1 minus the WHR on September 1.

[0065] Calculate basal metabolic rate BM: BM = BMR * 0.2389

[0066] The patient's basal metabolic rate (BMR) was calculated based on age, height and weight (Table 5).

[0067] BMR is expressed in kilojoules. To convert to kilocalories, multiply by 0.2389.

[0068] Age / years male female <18

[16] 370+20H+52BW-25A 1873+13H+39BW-18A 18～<30

[15] 63BW+2896 62BW+2036 30～60

[15] 48BW+3653 34BW+3538 >60

[16] 370+20H+52BW-25A 1873+13H+39BW-18A

[0069] Table 5: Calculation formula for BMR of people in different age groups

[0070] To calculate occupational consumption OM, energy consumption at rest should be equal to basal metabolic rate, while energy consumption during sitting, light activity, moderate physical activity, heavy physical activity and extremely heavy physical activity is approximately 1.2, 1.5, 4, 6 and 10 times of basal metabolic rate respectively.

[0071] Very slight activity: OM = 0.2*8*(BM / 24)

[0072] Light activity: OM = 0.5*8*(BM / 24)

[0073] Moderate physical activity: OM = 3*8*(BM / 24)

[0074] Heavy physical activity: OM = 5*8*(BM / 24)

[0075] Extreme physical activity: OM = 9*8*(BM / 24)

[0076] The calories consumed by exercise MM and the calories supplemented by diet DP are provided by the plan.

[0077] Data standardization: Take the base 10 logarithm of BM, OM, MM and then make negative values, take the base 10 logarithm of DP, multiply cWHR by 100, and cFPG does not need to be changed, so that these parameter values ​​are kept in a similar range that is easy to calculate and process, that is:

[0078]

[0079]

[0080]

[0081]

[0082] Wlq L =cWHR×100

[0083] Initialize model parameters: Initialize the weights w by sampling random numbers from a normal distribution with mean 0 and standard deviation 0.01, and initialize the bias value k to 0.

[0084] Define the model: Input feature x j and model weights w j The matrix-vector multiplication is followed by a bias value k.

[0085]

[0086] Define the data set input: x = [BM L ,OM L ,MM L ,DP L ] T

[0087] Define data set output: y = [cWHR L ] T .

[0088] Define the dataset weight: w = [w BM ,w OM ,w MM ,w DP ] T ,

[0089] Scalar deviation value: k

[0090] Among them [BM L ,OM L ,MM L ,DP L ] T For [BM L ,OM L ,MM L ,DP L ], w BM represents the basal metabolic weight, w OM Indicates occupational consumption weight, w MM represents the weight of exercise consumption, w DP represents the weight of dietary calorie supplement, and k represents the deviation value.

[0091] Define the loss function: When used to update the model, calculate the gradient of the loss function. When the gradient decreases, the weights should be updated in the opposite direction of the gradient, which can effectively find the global optimal solution.

[0092] Add the bias value to the weight: w←[w,k]

[0093] Calculate the square loss: e(x,y,w)=1 / 2n||y-xw|| 2

[0094] Among them, n means that there are n samples in the matrix. After the model obtains the data weights and bias values, it is necessary to learn and train according to the model. That is to say, the data needs to be included for updating and training to obtain the optimal equation coefficients. This n is the number of samples we will include later.

[0095] Define the optimization algorithm: Use mini-batch stochastic gradient descent as the optimization algorithm. When traversing the model, randomly extract mini-batches from the dataset and then calculate the gradient of the loss. The following function implements mini-batch stochastic gradient descent update. This function accepts the model parameter set x, y, w, the learning rate and batch size b as input. The size of each update step is given by Decide.

[0096]

[0097] Perform training: set the batch size b to 0.2 of the dataset length and set the learning rate is 0.01. After traversing 1000 times, when e is less than 0.1, the result is judged to be valid, and the data set weight w and scalar deviation value k are output. When k is greater than 0.1, the result is judged to be invalid, and the data set weight w and scalar deviation value k are not output. The nutritionist will formulate a diet plan.

[0098] The evaluation module is configured to automatically calculate the dietary supplement calorie value based on the optimal weights, deviation values, and gradient adjustment targets described in the calculation module, so as to generate a personalized diet plan and provide timely feedback for optimization.

[0099]

[0100]

[0101] Table 6: Results of BMI, waist circumference, waist-to-hip ratio, and waist-to-height ratio for each age group and gender

[0102] The principle is to adjust the patient's WHR to the average value of that age. For example, for a 52-year-old male patient, the average WHR value of 0.925 is set as the target waist-to-hip ratio pWHR (Table 6).

[0103] Define the current adjusted waist-to-hip ratio difference as:

[0104]

[0105] Among them, npWHR is the current adjusted waist-hip ratio difference, 0.005 is the maximum adjustment difference of waist-hip ratio in each period, nWHR is the latest waist-hip ratio, and pWHR is the target waist-hip ratio.

[0106] Define the current basal metabolic rate NBM, occupational calorie consumption NOM, exercise calorie consumption NMM, and dietary calorie supplement NDP, and take the logarithm of NBM, NOM, and NMM with a base of 10 and then negative it, take the logarithm of NDP with a base of 10, and multiply npWHR by 100, so that the values ​​of these parameters are all between -10 and 10.

[0107]

[0108]

[0109]

[0110]

[0111] nW L =npWHR×100

[0112] Calculate the calories added to the current diet plan:

[0113]

[0114] Calculate the weight of each type of food based on the food exchange method, and give priority to low GL foods: (GL stands for glycemic load)

[0115] Studies have shown that blood sugar levels after meals are closely related to the total amount of carbohydrates in food in addition to the GI (glycemic index) of carbohydrates. For foods with high GI, if there are few carbohydrates, although they are converted into blood sugar, they have little effect on blood sugar levels. If you choose a diet based solely on food GI and food calories, there may be deviations, because some foods have high GI but their GL is not necessarily high. Therefore, the present invention preferentially selects GL as a recommended solution for diabetic diet guidance (Table 7).

[0116]

[0117]

[0118]

[0119]

[0120] Table 7: Recommended diet plan for diabetes patients based on GL

[0121] The present invention also has embodiment 2 set according to individual actual situation:

[0122] Application implementation case (interconnection with APP): Patient A, male, 56 years old, height 170cm, weight 80kg, waist circumference 105cm, hip circumference 110cm, fasting blood glucose 9.0mmol / L, professional accountant.

[0123] Step 1: The client collects data related to the patient's energy consumption and energy replenishment

[0124] Step 1: A logs in to the collection and storage module for the first time. After successfully authorizing WeChat login, he supplements his name and phone number. The server uses SQL statements to query whether there is health assessment data for the patient in the database based on the patient's name and phone number. If not, the health assessment interface pops up.

[0125] Step 2: The patient fills out a health assessment, including gender, date of birth, height, weight, waist circumference, hip circumference, and work intensity.

[0126] Step 3: After completing the form, the patient clicks on Generate Diet Plan.

[0127] Step 4: The patient regularly updates height, weight, waist circumference, hip circumference, fasting blood glucose, and work intensity.

[0128] Step 5: After completing the daily exercise plan (if any) and diet plan, provide feedback on the actual work situation, actual exercise volume, and actual diet volume for the day.

[0129] Step 6: The collection and storage module sends data to the calculation module in a table format, which is empty if there is no exercise plan.

[0130] Step 7: The calculation module program determines whether the unit and format of the patient's data are correct. If correct, it proceeds to the next step. If incorrect, it returns the corresponding error reason and the collection and storage module reminds the patient to resubmit the data.

[0131] Step 2: The collection and storage module prepares a patient-specific data set based on the patient ID

[0132] Step 8: Data calculation: The calculation module program uses SQL statements to query all historical data of the patient in the database based on the patient ID, including gender, age, height, weight, waist circumference, hip circumference, fasting blood sugar, work intensity, exercise plan (if any) and diet plan.

[0133] According to the patient's WHR>0.9, the following is an example of waist-to-hip ratio difference adjustment:

[0134] Calculate the patient's WHR value (waist-to-hip ratio, waist / hip circumference) and sort them by measurement date. Traverse all WHR values ​​and make a difference. The difference defines cWHR. When the difference is greater than 0.01, define the time period as a single data set, and calculate BM, OM, MM, DP and cWHR in the time period. At the same time, store the data set for direct call next time.

[0135] Step 9: Standardize the data set, that is, take the base 10 logarithm of BM, OM, and MM and then negate them, take the base 10 logarithm of DP, and multiply cWHR by 100, so that the values ​​of these parameters are between -10 and 10.

[0136] Step 3: The collection and storage module imports the data set into the linear regression model to calculate the optimal weights of BM, OM, MM and DP

[0137] Step 10: The collection and storage module inputs all the patient data sets into the linear regression model. The model calculates the weights and scalar deviations of the basal metabolic rate, occupational calorie consumption, exercise calorie consumption, and dietary calorie supplementation, then adds the deviation value to the weight and calculates the least square loss to obtain the optimal weights for each item. The dietary calorie supplementation for the current diet plan is then calculated based on the weights.

[0138] Step 11: Define the data set input: x = [BM L ,OM L ,MM L ,DP L ] T .

[0139] Step 12: Define the data set output: y = [cWHR L ] T .

[0140] Step 13: Define weights and scalar bias: w = [w BM ,w OM ,w MM ,w DP ] T , k.

[0141] Step 14: Add the bias value to the weight: w←[w,b], and calculate the square loss: e(x,y,w)=1 / 2n||y-xw|| 2

[0142] Step 15: Set the batch size b to 0.2 of the dataset length and set the learning rate is 0.01. After traversing 1000 times, when e is less than 0.1, the result is judged to be valid, and the data set weight w and scalar deviation value k are output. When k is greater than 0.1, the result is judged to be invalid, and the data set weight w and scalar deviation value k are not output. The nutritionist will formulate a diet plan.

[0143] Step 16: The calculation result is

[0144] w BM =25.3373; w OM =5.9850; w MM =3.9536; w DP =25.2024; k=34.9801

[0145] Step 4: Set a gradient adjustment target for WHR (waist-to-hip ratio) based on the patient's gender and age, dynamically evaluate and automatically push dietary calorie supplementation recommendations

[0146] Step 17: Based on gender A, male, age 56, the pWHR (average waist-to-hip ratio) value is 0.521

[0147] Step 18: Calculate the current adjusted waist-to-hip ratio difference:

[0148]

[0149] We get npWHR = -0.05

[0150] Step 19: Set the current (30 days) basal metabolic rate NBM to 51780kal; the current occupational consumption NOM to 8640cal; and the current exercise consumption NMM to 12000cal.

[0151] Step 20: Calculate the calories needed for the current diet:

[0152]

[0153] The NDP is 79432 cal.

[0154] Step 21: Push the calculation results obtained by the calculation module to the collection and storage module to display them to the patient.

[0155] Step 4: Recommend low GL foods

[0156] Step 22: The collection and storage module calculates the weight of each type of food according to the food exchange portion method, and gives priority to recommending low-GL food.

[0157] The case operation shows that the present invention has the following obvious effects:

[0158] (1) The present invention establishes an individualized data set by collecting energy consumption data of diabetic patients in an individualized manner in the early stage, and then establishes an improved linear regression model. The dual-gradient target is set by using the waist-to-hip ratio (WHR), which is a sensitive indicator of the patient's visceral fat content, and the change in the mean blood glucose level (FPG), so as to more accurately derive the patient's calorie supplement plan and automatically generate a diet plan in a timely manner.

[0159] (2) The present invention can combine the patient's personal physical data, occupation, and lifestyle habits such as exercise and diet, accurately grasp the patient's calorie consumption characteristics through program learning and training, and then automatically adjust the patient's diet plan through timely feedback.

[0160] (3) The present invention targets WHR and FPG, which can better improve important physiological indicators such as related visceral fat, body weight, waist circumference, hip circumference, fasting blood glucose, etc., and more accurately and quickly achieve disease remission in diabetic patients.

[0161] (4) The present invention selects BM (basal metabolic rate), OM (occupational calories), MM (exercise calories), and DP (dietary calories) related to WHR and FPG as the main basis for modeling, and dynamically calculates the weights of the four indicators through automatic learning through a program. This not only more accurately reflects the actual energy consumption and dietary supplementation of patients, but is also simpler and more practical than traditional dietary guidance methods.

[0162] (5) The data processing of the present invention is based on the improved linear regression model. The collection and analysis of relevant indicators can achieve dynamic feedback and the push of dietary guidance simultaneously, which greatly simplifies the cumbersome clinical dietary plan formulation process and improves the efficiency of patient dietary management.

[0163] (6) The optimal caloric supplement value obtained by the present invention after deep learning can provide a basic basis for adjusting the patient's carbohydrate, protein and fat intake and proportion in combination with the patient's fasting blood glucose level fluctuations.

[0164] It is understandable that the related features in the above system can be referenced to each other. In addition, the "first", "second" and so on in the above embodiments are used to distinguish the various embodiments, and do not represent the advantages and disadvantages of the various embodiments.

[0165] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0166] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0167] In addition, the memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0168] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0169] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0170] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0171] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0172] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0173] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0174] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0175] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0176] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0177] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

[0178] It should be noted that the above embodiments do not limit the present invention in any form, and any technical solutions obtained by equivalent replacement or equivalent transformation shall fall within the protection scope of the present invention.

Claims

1. A deep learning-based adaptive generation system for diabetes diet plans, characterized in that: include: Collection and storage module, The system is configured to collect and generate data on the dynamic changes in the total calories of patients, and to construct a personalized data set that can be used for machine learning training and generate a calorie supplement plan based on the basal metabolic rate, occupational consumption, exercise consumption, dietary supplements, and the difference between the waist-to-hip ratio and the average fasting blood sugar; A calculation module, configured as an improved linear regression model, imports the personalized data set into the model and performs data training by a program to obtain the optimal weight corresponding to each data in the data set; The evaluation and feedback module is configured to automatically calculate the dietary supplement calorie value based on the optimal weight and deviation value described in the calculation module by setting a dual gradient target of waist-to-hip ratio difference and fasting blood glucose mean difference, so as to generate a personalized diet plan and provide timely feedback for optimization; The improved linear regression model is: the output feature value is y = input feature x j and model weights w j After matrix-vector multiplication + bias value k, that is: Among them, the output eigenvalue y is the double gradient target value. When the male waist-to-hip ratio WHR m >0.9, female WHR f When >0.85, y is the waist-to-hip ratio gradient target value cWHR, and the input eigenvalue x i is the value of each data in the data set, basal metabolic rate, occupational consumption, exercise consumption, and dietary supplement, and the model weight w i is the weight of the ith item; When the waist-to-hip ratio (WHR) of men m ≤0.9, female waist-to-hip ratio WHR f When ≤0.85, y is the fasting blood glucose gradient target value cFPG, and the input characteristic value x j is the value of each data in the data set, basal metabolic rate, occupational consumption, exercise consumption, and dietary supplement, and the model weight w j is the j-th weight; The relevant data include BM basal metabolic rate, OM occupational calorie consumption, MM exercise calorie consumption and DP dietary calorie supplementation, which together with the established cWHR and cFPG gradient target values ​​constitute a complete personalized data set for program training; Among them, basal metabolic mass BM = BMR*0.2389, BMR is basal metabolic rate, which is calculated based on age, height and weight; occupational consumption OM = h*8*(BM / 24), h is the energy consumption of very light activity, light activity, moderate physical activity, heavy physical activity and very heavy physical activity represented by different occupations of individuals, which are 1.2, 1.5, 4, 6 and 10 times of the basal metabolic rate respectively.

2. A diabetes diet plan adaptive generation system based on deep learning according to claim 1, characterized in that: The collection and storage module is configured to dynamically collect and update the relevant data of the patient's total calorie consumption, continuously form new data sets, and transmit the new data sets to the calculation module and the evaluation and feedback module in sequence for processing, and push the dynamically updated calorie supplement plan according to the instant optimal weight; When dynamically updating the heat supplementation plan, there is a gradient of the loss function. When the gradient decreases, the weight is updated immediately in the opposite direction of the gradient. The weights w are obtained according to the dual gradient target background. i 、w j And the optimal solutions for the scalar deviation value k include: Add the bias value to the weight: w←[w,k]; Calculate the square loss: e(x, y, w) = 1 / 2n||y-xw|| 2 ; n means there are n samples in the matrix, the square loss e is the error between the sample prediction value and the actual value. The smaller the value, the smaller the loss. w is the matrix form of the weight.

3. The deep learning-based adaptive generation system for diabetes diet according to claim 2, characterized in that: Set the batch size b as the length of the patient's personalized data set, set l as the learning rate, and after traversing the set number of times, when e is less than the set threshold, the judgment result is valid, and the data set weight w and scalar deviation value k are output. When k is greater than the set threshold, the judgment result is invalid. The function accepts the model parameter set x, y, w, l and b as input, and the size of each step update is determined by l; 4. The deep learning-based adaptive generation system for diabetes diet plans according to claim 3, characterized in that: The process also includes a data standardization process, where BM, OM, and MM are taken as base 10 logarithms and then negative, DP is taken as base 10 logarithms, cWHR is multiplied by 100, and cFPG does not need to be changed, so that these parameter values ​​are kept in a similar range that is convenient for calculation and processing, that is: cWHR L =cWHR×100 The waist-to-hip ratio WHR=W / H, and cWHR is the difference in waist-to-hip ratios at different time points.

Images