Diabetic foot incidence prediction model construction method and system based on deep learning

Abstract

The invention discloses a deep learning-based diabetic foot incidence prediction model construction method and system, and relates to the technical field of diabetic foot prediction, and the method comprises the following steps: collecting foot thermal imaging information and basic physiological information of a diabetic patient, and constructing a patient foot reference data set; performing first temperature processing to obtain foot temperature difference change data; feature marking processing is carried out to obtain foot temperature sample data, and a diabetic foot basic prediction model is constructed; first similar patients are screened, a diabetic foot individuality prediction model is constructed according to foot temperature sample data of the first similar patients, and diabetic foot onset risk prediction is carried out; the method is used for solving the problem that in an existing diabetic foot prediction technology, when prediction is carried out through foot temperature, personalized diabetic foot attack risk prediction cannot be carried out on a diabetic patient according to temperature difference changes of feet of the diabetic patient and basic physiological information of the patient.

Description

Technical Field

[0001] The present invention relates to the technical field of diabetic foot prediction, and specifically to a method and system for constructing a diabetic foot onset prediction model based on deep learning. Background Art

[0002] The diabetic foot prediction technology refers to a series of methods and means that comprehensively utilize knowledge in multiple fields such as medicine, information technology, and data science, collect and analyze various data related to diabetic patients, and construct a model to estimate the possibility of an individual developing diabetic foot within a specific future time period.

[0003] When the existing diabetic foot prediction technology predicts through foot temperature, it often measures the temperatures of different parts of the foot and sets a fixed temperature difference threshold to judge the onset risk of diabetic foot. For example, when measuring the temperature of the dorsal foot and the sole of the foot, if the temperature difference exceeds 4°C, it is considered to have a relatively high onset risk. However, it ignores the time-varying characteristics of the foot temperature difference. There may be many important details in the change process of the temperature difference, such as the change frequency, amplitude, and acceleration. These details can reflect the rapid or progressive changes in the foot's physiological state. For example, if the temperature difference rapidly increases or decreases within a short period of time, it may imply sudden vasoconstriction or dilation of the foot blood vessels, rapid changes in nerve function, etc., which are often important warning signals for the onset of diabetic foot. Merely focusing on the static temperature difference will cause these key dynamic change details to be ignored, and it is impossible to timely and accurately detect the urgent changes in the condition. Moreover, due to large individual physiological differences, relying solely on a single static temperature difference lacks flexibility. Different people have different basal body temperatures and foot temperatures, which may lead to misjudgment. For example, in the patent application with the publication number CN113096811A, a diabetic foot image processing and risk warning device based on infrared thermal imaging, this solution conducts risk warning by obtaining the infrared thermal imaging of the foot, but ignores the time-varying characteristics of the infrared thermal imaging of the foot and cannot be optimized individually according to patients, thus reducing the accuracy and reliability of the risk warning. Therefore, when the existing diabetic foot prediction technology predicts through foot temperature, it cannot conduct personalized diabetic foot onset risk prediction for diabetic patients based on the change of the foot temperature difference of diabetic patients and their own basic physiological information. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems in the prior art to some extent. By collecting the foot thermal imaging information and basic physiological information of diabetic patients, a reference data set of the patients' feet is constructed; and a first temperature processing is performed to obtain the data of the change in foot temperature difference; according to the data of the change in foot temperature difference, a feature marking process is carried out to obtain the foot temperature sample data, and a basic prediction model for diabetic foot is constructed; the first similar patients are screened, and a personalized prediction model for diabetic foot is constructed based on the foot temperature sample data of the first similar patients, and the risk prediction of diabetic foot onset is carried out, so as to solve the problem that the existing diabetic foot prediction technology cannot perform personalized risk prediction of diabetic foot onset for diabetic patients based on the change in foot temperature difference and the basic physiological information of the patients themselves when predicting through foot temperature.

[0005] To achieve the above object, in the first aspect, the present application provides a method for constructing a diabetic foot onset prediction model based on deep learning, including the following steps:

[0006] Collect the foot thermal imaging information and basic physiological information of diabetic patients, and construct a reference data set of the patients' feet;

[0007] Perform a first temperature processing on the foot thermal imaging information of diabetic patients to obtain the data of the change in foot temperature difference;

[0008] According to the data of the change in foot temperature difference, perform a feature marking process to obtain the foot temperature sample data, and construct a basic prediction model for diabetic foot;

[0009] According to the basic physiological information of diabetic patients, screen the first similar patients, and construct a personalized prediction model for diabetic foot based on the foot temperature sample data of the first similar patients, and perform the risk prediction of diabetic foot onset.

[0010] Further, collecting the foot thermal imaging information and basic physiological information of diabetic patients and constructing a reference data set of the patients' feet includes the following steps:

[0011] Select the key parts of the foot, denoted as R1 to Rn respectively, and record the actual positions of R1 to Rn on the foot; and set the first acquisition method and the second acquisition method;

[0012] The first acquisition method includes: keeping the diabetic patient in a quiet state, fully exposing the foot of the diabetic patient, and after the foot temperature is stable, for any one of the key parts of the foot, use a thermal imaging device to directly face the key part of the foot to be measured for thermal imaging shooting to obtain a static thermal imaging image;

[0013] The second acquisition method includes: fully exposing the feet of a diabetic patient, and after allowing the diabetic patient to walk normally for t0 minutes, for any one of the key foot parts of the foot, a thermal imaging device is used to directly face the key foot part to be measured for thermal imaging shooting to obtain a dynamic thermal imaging image.

[0014] Furthermore, collecting the thermal imaging information and basic physiological information of the feet of diabetic patients, and constructing a patient foot reference dataset includes the following sub-steps:

[0015] For any diabetic patient, obtain the gender, age, height, weight, and BIM of the diabetic patient, and at a first time interval, use the first acquisition method and the second acquisition method to obtain the static thermal imaging images R1 to Rn and the corresponding dynamic thermal imaging images of the diabetic patient, and record the acquisition time and the number of acquisitions at the same time. Denote the static thermal imaging image and the dynamic thermal imaging image collected at the same time as a static-dynamic thermal imaging group, and sort them according to the acquisition time, which is denoted as the thermal imaging information of the feet of the diabetic patient;

[0016] While obtaining the static thermal imaging image of the diabetic patient each time, collect the body temperature and heart rate of the diabetic patient, sort them according to the acquisition time, and store them together with the gender, age, height, weight, and BIM of the diabetic patient, which is denoted as the basic physiological information of the diabetic patient;

[0017] Obtain the thermal imaging information and basic physiological information of the feet of multiple diabetic patients, and classify them according to the corresponding patients, which is denoted as the patient foot reference dataset.

[0018] Furthermore, perform a first temperature processing on the thermal imaging information of the feet of diabetic patients to obtain foot temperature difference change data, including the following sub-steps:

[0019] For any thermal imaging image in the thermal imaging information of the feet of any diabetic patient, perform temperature correction extraction processing. The temperature correction extraction processing includes: for any thermal imaging image, denoted as the first foot thermal imaging, divide the thermal imaging area corresponding to the key foot part in the first foot thermal imaging, which is denoted as the key part thermal imaging area; obtain the maximum value and the minimum value of the temperature values corresponding to all pixel points in the key part thermal imaging area, and denote them as Tma and Tmi in sequence; set the interval size as b, and divide the interval into multiple sub-intervals of size b, which are denoted as basic temperature sub-intervals;

[0020] For any pixel point in the key part thermal imaging area, denoted as the first central pixel point; set the neighborhood range size as a*a, and with the first central pixel point as the center, obtain a neighborhood pixel area of size a*a, which is denoted as the first central neighborhood;

[0021] Obtain the temperature values corresponding to all pixel points in the first central neighborhood, count the basic temperature sub-intervals to which all temperature values belong respectively, and then count the basic temperature sub-interval with the most occurrences, denoted as the majority temperature interval. Determine whether the basic temperature sub-interval corresponding to the corresponding first central pixel point belongs to the majority temperature interval. If it does not belong, and the ratio of the number of majority temperature intervals to the total number of basic temperature sub-intervals corresponding to the first central neighborhood exceeds K0, then mark the first central pixel point as an abnormal pixel point; otherwise, mark the first central pixel point as a normal pixel point. Repeat marking all pixel points in the thermal imaging area of the key part;

[0022] Remove all abnormal pixel points in the thermal imaging area of the key part after marking is completed, obtain the temperature values corresponding to all normal pixel points, then calculate the average temperature value, and mark it as the corresponding static temperature value or dynamic temperature value according to the static thermal imaging image or dynamic thermal imaging image corresponding to the thermal imaging area of the key part.

[0023] Further, the first temperature processing of the foot thermal imaging information of diabetic patients to obtain the foot temperature difference change data also includes the following sub-steps:

[0024] Repeatedly obtain all static temperature values or dynamic temperature values in all foot thermal imaging information, and calculate the difference between the static temperature value and the dynamic temperature value corresponding to the same acquisition time of the same foot key part according to the first temperature difference formula, denoted as the static-dynamic temperature difference. The first temperature difference formula is as follows: R0 = AR - BR, where R0 represents the static-dynamic temperature difference, AR represents the static temperature value, and BR represents the corresponding dynamic temperature value;

[0025] Obtain all corresponding static-dynamic temperature differences in all foot thermal imaging information. For any foot thermal imaging information, classify all corresponding static-dynamic temperature differences according to the foot key parts to which they belong, and sort them according to the corresponding acquisition time and acquisition times, denoted as foot temperature difference change information;

[0026] Repeatedly obtain the foot temperature difference change information of all patients, denoted as foot temperature difference change data.

[0027] Further, perform feature marking processing according to the foot temperature difference change data to obtain foot temperature sample data, and construct a basic diabetic foot prediction model, including the following sub-steps:

[0028] According to the foot temperature difference change data of the same key foot parts of the same diabetic patient and the corresponding basic physiological information, the dynamic and static temperature difference, heart rate, and body temperature of the same collection time are used to form a risk feature vector, denoted as M = {m1, m2, m3}, where m1, m2, and m3 represent the dynamic and static temperature difference, heart rate, and body temperature in sequence, respectively. The risk feature vector is sorted according to the corresponding collection times, denoted as the risk feature data of the corresponding key foot parts;

[0029] Obtain the risk feature data of all key foot parts of all diabetic patients, and mark them according to whether the corresponding diabetic patients have diabetic foot, denoted as foot temperature sample data.

[0030] Furthermore, feature marking processing is performed on the foot temperature difference change data to obtain foot temperature sample data, and the basic prediction model for diabetic foot also includes the following sub-steps:

[0031] Based on constructing the original prediction model, the original prediction model includes an input layer, an LSTM layer, a fully connected layer, and an output layer. Set the number of neurons in the input layer as c1, the number of neurons in the LSTM layer as c2, the number of neurons in the fully connected layer as c3, and the number of neurons in the output layer as c4. And set the activation function of the output layer as the sigmoid activation function, and use the foot temperature sample data to train the original prediction model. After completion, the basic prediction model is obtained.

[0032] Furthermore, screening the first similar patients according to the basic physiological information of diabetic patients includes the following sub-steps:

[0033] For the diabetic patient to be predicted, denoted as the patient to be tested, obtain the gender, age, height, weight, and BIM of the patient to be tested, and set the age similarity threshold as e1, the height similarity threshold as e2, the weight similarity threshold as e3, and the BIM similarity threshold as e4;

[0034] Mark other diabetic patients with the same gender as the patient to be tested and who simultaneously satisfy that the age difference is less than or equal to e1, the height difference is less than or equal to e2, the weight difference is less than or equal to e3, and the BIM difference is less than or equal to e4 as the suspected similar patients of the patient to be tested; Obtain the suspected similar patients of all patients to be tested in the patient foot reference dataset;

[0035] Obtain the foot thermal imaging information and basic physiological information of the key foot parts of the patient to be tested at the first time interval; Denote them as the to-be-tested thermal imaging information and the to-be-tested physiological information in sequence, and calculate the average heart rate and average body temperature of the patient to be tested according to the to-be-tested physiological information;

[0036] According to the patient foot reference dataset, obtain the average heart rate and average body temperature of all corresponding suspected similar patients;

[0037] Among all the suspected similar patients of the patient to be tested, the suspected similar patients who simultaneously satisfy that the difference in the average heart rate from the patient to be tested is less than f1 and the difference in the average body temperature is less than f2 are marked as the first similar patients, where f1 is the set heart rate similarity threshold and f2 is the set body temperature similarity threshold.

[0038] Furthermore, a diabetic foot personality prediction model is constructed based on the foot temperature sample data of the first similar patients, and the prediction of the onset risk of diabetic foot includes the following sub-steps:

[0039] Obtain the foot temperature sample data corresponding to the first similar patients, denoted as personality sample data, and use the personality sample data to train the basic prediction model. After completion, obtain the diabetic foot personality prediction model of the basic prediction model;

[0040] Perform the first temperature processing on the thermal imaging information to be tested of the patient to be tested, and perform feature marking processing. Then, input it into the diabetic foot personality prediction model according to the corresponding key foot parts respectively to obtain the onset risk probability of the selected key foot parts.

[0041] In a second aspect, the present application provides a system for constructing a diabetic foot onset prediction model based on deep learning, including an information collection module, an information processing module, a basic construction module, and a risk prediction module;

[0042] The information collection module is used to collect the foot thermal imaging information and basic physiological information of diabetic patients, and construct a reference data set of patients' feet;

[0043] The information processing module is used to perform the first temperature processing on the foot thermal imaging information of diabetic patients to obtain foot temperature difference change data;

[0044] The basic construction module is used to perform feature marking processing according to the foot temperature difference change data to obtain foot temperature sample data, and construct a basic diabetic foot prediction model;

[0045] The risk prediction module includes a strengthening unit and a prediction unit. The strengthening unit screens the first similar patients according to the basic physiological information of diabetic patients, and constructs a diabetic foot personality prediction model according to the foot temperature sample data of the first similar patients. The prediction unit is used to predict the onset risk of diabetic foot.

[0046] Advantages of the present invention: By collecting the foot thermal imaging information and basic physiological information of diabetic patients, a reference data set of the patients' feet is constructed; the foot thermal imaging information of diabetic patients is subjected to a first temperature processing to obtain foot temperature difference change data; feature marking processing is performed according to the foot temperature difference change data to obtain foot temperature sample data, and a basic diabetic foot prediction model is constructed; the first similar patients are screened according to the basic physiological information of diabetic patients, and a personalized diabetic foot prediction model is constructed according to the foot temperature sample data of the first similar patients, and the onset risk of diabetic foot is predicted. It is possible to perform personalized prediction of the onset risk of diabetic foot for diabetic patients based on the temperature difference change of the diabetic patients' feet and the patients' own basic physiological information;

[0047] The present invention collects the foot temperature difference change data of diabetic patients. The advantages are that many diabetic feet may not cause obvious temperature differences in the early stage, but will cause subtle changes in temperature differences, which helps to detect problems in a timely manner in the early stage of the disease, and can exclude the influence of environmental and other interference factors, making the risk prediction more accurate; by screening the temperature intervals of the key part thermal imaging areas, compared with the traditional method, outliers can be simply removed, and the effective data features can be retained, making the subsequent temperature difference calculation more accurate and providing a more reliable data basis for model construction; by using the data of similar patients to perform secondary training on the constructed model, the advantage is that it can better adapt to the influence of the individual characteristics of different patients on the onset risk of diabetic foot, thereby improving the accuracy and pertinence of the prediction and avoiding misjudgment caused by individual differences. Brief Description of the Drawings

[0048] Figure 1 It is the principle block diagram of the system of the present invention;

[0049] Figure 2 It is the step flow chart of the method of the present invention;

[0050] Figure 3 It is the first central neighborhood schematic diagram of the present invention;

[0051] Figure 4 It is the original prediction model structure diagram of the present invention;

[0052] Figure 5 It is the structure schematic diagram of the electronic device of the present invention. Detailed Embodiments

[0053] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] Example 1. Refer to Figure 1 As shown, the present application provides a system for constructing a diabetes foot onset prediction model based on deep learning, including an information collection module, an information processing module, a basic construction module, and a risk prediction module;

[0055] The information collection module is used to collect the foot thermal imaging information and basic physiological information of diabetic patients, and construct a reference data set of the patients' feet;

[0056] The information collection module is configured with an information collection strategy, which includes: selecting key parts of the foot, denoted as R1 to Rn respectively, and recording the actual positions of R1 to Rn on the foot; and setting a first collection method and a second collection method; the key parts of the foot include toes, dorsal foot, sole, inner and outer sides of the ankle, etc.

[0057] The first collection method includes: keeping the diabetic patient in a quiet state, fully exposing the foot of the diabetic patient, and after the foot temperature stabilizes, for any one of the key parts of the foot, taking a thermal imaging shot of the key part of the foot to be measured through a thermal imaging device to obtain a static thermal imaging image;

[0058] The second collection method includes: fully exposing the foot of the diabetic patient and letting the diabetic patient walk normally for t0 minutes, and then for any one of the key parts of the foot, taking a thermal imaging shot of the key part of the foot to be measured through a thermal imaging device to obtain a dynamic thermal imaging image; t0 can be set according to the actual application scenario, but it is necessary to ensure that t0 is the same for different patients, generally 5 to 10 minutes;

[0059] For any diabetic patient, obtain the gender, age, height, weight, and BMI of the diabetic patient, and at a first time interval, use the first collection method and the second collection method to obtain the static thermal imaging images and corresponding dynamic thermal imaging images of R1 to Rn of the diabetic patient, and record the collection time and the number of collections at the same time. Denote the static thermal imaging image and the dynamic thermal imaging image collected at the same time as a static-dynamic thermal imaging group, and sort them according to the collection time, denoted as the foot thermal imaging information of the diabetic patient. The first time interval is t1; for one collection of a key part of the foot, it is necessary to collect a static thermal imaging image, a dynamic thermal imaging image, body temperature, and heart rate; in this embodiment, the first time interval t1 is 24 hours;

[0060] While acquiring the static thermal imaging image of a diabetic patient each time, the body temperature and heart rate of the diabetic patient are collected, sorted according to the collection time, and stored in combination with the gender, age, height, weight, and BMI of the diabetic patient, which is recorded as the basic physiological information of the diabetic patient;

[0061] Obtain the foot thermal imaging information and basic physiological information of multiple diabetic patients, and classify them according to the corresponding patients, which is recorded as the patient foot reference dataset;

[0062] In the specific implementation process, diabetes often causes vascular lesions, affecting foot blood circulation; normally, blood brings heat to tissues to maintain body temperature; when blood vessels are stenosed or blocked, blood perfusion in the corresponding area decreases and the temperature drops; after walking, the temperature change in the area where the sole pressure is concentrated is abnormal, which may be due to blood vessel compression or lesions, resulting in changes in blood supply and heat dissipation, reflecting the relationship between blood circulation status and the risk of diabetic foot. By comparing the temperature difference between rest and walking, the dynamic changes of foot blood circulation and metabolic function during physical activity can be understood. For example, when walking compared to rest, the temperature change in some areas is large, indicating that when the body posture changes, the ability of the foot to cope with blood flow and metabolic needs is abnormal, which is related to the risk of diabetic foot.

[0063] The information processing module is used to perform the first temperature processing on the foot thermal imaging information of diabetic patients to obtain foot temperature difference change data;

[0064] The information processing module is configured with an information processing strategy, and the information processing strategy includes: performing temperature correction and extraction processing on any thermal imaging image in the foot thermal imaging information of any diabetic patient. The temperature correction and extraction processing includes: for any thermal imaging image, denoted as the first foot thermal imaging, dividing the thermal imaging area corresponding to the key parts of the foot in the first foot thermal imaging, denoted as the key part thermal imaging area; obtaining the maximum and minimum values of the temperature values corresponding to all pixel points in the key part thermal imaging area, denoted as Tma and Tmi in sequence; setting the interval size as b, and dividing the interval into multiple sub-intervals with a size of b, denoted as basic temperature sub-intervals; in this embodiment, b = 2. For example, if Tma = 35.3 °C and Tmi = 30.2 °C, then the interval is [30, 36], and the basic temperature sub-intervals are [30, 32), [32, 34), and [34, 36];

[0065] For any pixel point in the key part thermal imaging area, denoted as the first central pixel point; Please refer to Figure 3As shown, set the neighborhood range size to a*a. Centered on the first central pixel point, obtain a neighborhood pixel area of a*a size, denoted as the first central neighborhood. In this embodiment, a*a = 5*5, and a is generally an odd number, that is, a 5*5 pixel point area is the first central neighborhood. For edge pixel points, if their first central neighborhood does not have a*a pixel points, subsequent processing is performed according to the actual existing pixel points.

[0066] Obtain the temperature values corresponding to all pixel points in the first central neighborhood, and count the basic temperature sub-intervals to which all temperature values belong respectively. Then count the basic temperature sub-interval with the most occurrences, denoted as the majority temperature interval. Determine whether the basic temperature sub-interval corresponding to the first central pixel point belongs to the majority temperature interval. If not, and the ratio of the number of majority temperature intervals to the total number of basic temperature sub-intervals corresponding to the first central neighborhood exceeds K0, then mark the first central pixel point as an abnormal pixel point; otherwise, mark the first central pixel point as a normal pixel point. Repeat the marking for all pixel points in the thermal imaging area of the key part. In this embodiment, K0 = 0.6. For example, the basic temperature sub-intervals corresponding to a certain first central neighborhood are [30, 32), [32, 34), and [34, 36), and the corresponding temperature values are Then the corresponding basic temperature sub-interval is The majority temperature interval is [34, 36], and the basic temperature sub-interval corresponding to the first central pixel point is [30, 32), which does not belong to the majority temperature interval. And the number of majority temperature intervals is 16, 16 / 25 > 0.6, so the first central pixel point is an abnormal pixel point.

[0067] Remove all abnormal pixel points in the thermal imaging area of the key part after marking is completed, and obtain the temperature values corresponding to all normal pixel points. Then calculate the average temperature value, and mark it as the corresponding static temperature value or dynamic temperature value according to the static thermal imaging image or dynamic thermal imaging image corresponding to the thermal imaging area of the key part.

[0068] Repeat to obtain all static temperature values or dynamic temperature values in all foot thermal imaging information, and calculate the difference between the static temperature value and the dynamic temperature value corresponding to the same acquisition times of the same key part of the foot according to the first temperature difference formula, that is, the difference between the static temperature value and the dynamic temperature value corresponding to the static thermal imaging image and the dynamic thermal imaging image of the same part at the same acquisition. Denote it as the static-dynamic temperature difference. The first temperature difference formula is as follows: R0 = AR - BR, where R0 represents the static-dynamic temperature difference, AR represents the static temperature value, and BR represents the corresponding dynamic temperature value.

[0069] Obtain all corresponding dynamic and static temperature differences in all foot thermal imaging information. For any foot thermal imaging information, classify all corresponding dynamic and static temperature differences according to the key foot parts they belong to, and sort them according to the corresponding acquisition time and acquisition times, which is denoted as foot temperature difference change information;

[0070] Repeat to obtain the foot temperature difference change information of all patients, which is denoted as foot temperature difference change data;

[0071] In the specific implementation process, by screening the basic temperature sub-intervals of the key part thermal imaging areas, the advantage is that the outliers are judged based on the consistency of the central pixel and the neighborhood pixel features, which is easy to operate and implement. Through temperature interval quantization and majority interval judgment, abnormal pixel points can be effectively identified; the reliability of the judgment is ensured, and the overall temperature distribution characteristics of the image are retained to the greatest extent; the subsequent temperature difference calculation is more accurate, providing a more reliable data basis for model construction.

[0072] The basic construction module is used to perform feature marking processing on the foot temperature difference change data to obtain foot temperature sample data, and construct a basic prediction model for diabetic foot;

[0073] The basic construction module is configured with a basic construction strategy, and the basic construction strategy includes: according to the foot temperature difference change data of the same key foot part of the same diabetic patient and the corresponding basic physiological information, the dynamic and static temperature differences, heart rate and body temperature at the same acquisition time form a risk feature vector, which is denoted as M={m1, m2, m3}, where m1, m2 and m3 represent the dynamic and static temperature differences, heart rate and body temperature in sequence, and sort the risk feature vector according to the corresponding acquisition times, which is denoted as the risk feature data of the corresponding key foot part; forming the risk feature vector is convenient for subsequent model training; when forming the risk feature vector, the heart rate and body temperature are included because the heart rate and body temperature can reflect the patient's physical condition from different angles, and through comprehensive analysis by the deep learning model, the risk of diabetic foot can be evaluated more comprehensively and accurately;

[0074] Obtain the risk feature data of all key foot parts of all diabetic patients, and mark them according to whether the corresponding diabetic patients have diabetic foot, which is denoted as foot temperature sample data;

[0075] Based on constructing the original prediction model, please refer to Figure 4As shown in the figure, the original prediction model includes an input layer, an LSTM layer, a fully connected layer, and an output layer. The number of neurons in the input layer is set to c1, the number of neurons in the LSTM layer is set to c2, the number of neurons in the fully connected layer is set to c3, and the number of neurons in the output layer is set to c4. The activation function of the output layer is set to the sigmoid activation function. The sigmoid activation function can convert the output of the fully connected layer into a probability value between 0 and 1, representing the risk probability of diabetic foot onset. The original prediction model is trained using the foot temperature sample data, and after completion, a basic prediction model is obtained. In this embodiment, c1 = 3 because only the dynamic and static temperature difference, heart rate, and body temperature are input into the model, c2 = 128, c3 = 64, and c2 and c3 can be set according to the actual application scenario, c4 = 1, and a risk probability is output.

[0076] In the specific implementation process, the foot temperature sample data can be divided into a training set and a test set according to a certain ratio. For the completed basic prediction model, the test set can be used for testing, and indicators such as the accuracy, recall rate, F1 value, and precision rate of the model can be calculated to comprehensively evaluate the performance of the model. If the performance of the model does not meet the expectations, training can be carried out again.

[0077] The risk prediction module includes a strengthening unit and a prediction unit. The strengthening unit screens the first similar patients according to the basic physiological information of diabetic patients and constructs a personalized prediction model for diabetic foot based on the foot temperature sample data of the first similar patients. The prediction unit is used to predict the onset risk of diabetic foot.

[0078] The strengthening unit is configured with a model strengthening strategy. The model strengthening strategy includes: for the diabetic patient to be predicted, denoted as the patient to be measured, obtain the gender, age, height, weight, and BMI of the patient to be measured, and set the age similarity threshold to e1, the height similarity threshold to e2, the weight similarity threshold to e3, and the BMI similarity threshold to e4.

[0079] Other diabetic patients with the same gender as the patient to be measured and simultaneously satisfying that the age difference is less than or equal to e1, the height difference is less than or equal to e2, the weight difference is less than or equal to e3, and the BMI difference is less than or equal to e4 are marked as the suspected similar patients of the patient to be measured. Obtain all the suspected similar patients of the patients to be measured in the patient foot reference dataset. In this embodiment, e1 = 3 years old, e2 = 5 cm, e3 = 8 kg, e4 = 2. For example, patient one, 32 years old, 175 cm, 56 kilograms; BMI = 18.3 kg / m 2 , patient two, 35 years old, 178 cm, 60 kilograms; BMI = 18.5 kg / m 2 , then patient one and patient two are suspected similar patients to each other.

[0080] Obtain the foot thermal imaging information and basic physiological information of the key parts of the foot of the patient to be tested at the first time interval; record them as the thermal imaging information to be tested and the physiological information to be tested in sequence, and calculate the average heart rate and average body temperature of the patient to be tested according to the physiological information to be tested;

[0081] According to the patient foot reference data set, obtain the average heart rate and average body temperature of all corresponding suspicious similar patients; that is, calculate the average heart rate and average body temperature according to the collected heart rate and body temperature;

[0082] Among all the suspicious similar patients of the patient to be tested, mark the suspicious similar patients who simultaneously satisfy that the difference from the average heart rate of the patient to be tested is less than f1 and the difference from the average body temperature is less than f2 as the first similar patients, where f1 is the set heart rate similarity threshold and f2 is the set body temperature similarity threshold; in this embodiment, f1 = 10 times and f2 = 0.4 °C;

[0083] The prediction unit is configured with a risk prediction strategy, and the risk prediction strategy includes: obtaining the foot temperature sample data corresponding to the first similar patients, denoted as personalized sample data, and using the personalized sample data to train the basic prediction model. After completion, obtain the diabetes foot personalized prediction model of the basic prediction model;

[0084] Perform the first temperature processing on the thermal imaging information to be tested of the patient to be tested, and perform feature marking processing, and then input it into the diabetes foot personalized prediction model according to the corresponding key parts of the foot respectively to obtain the disease risk probability of the selected key parts of the foot;

[0085] In the specific implementation process, the physical conditions, living habits, disease progress, etc. of different diabetic patients are different. It is difficult to accurately adapt to each patient only using the basic prediction model; through the first similar patients, a group similar to the target patient in terms of physiological characteristics and other aspects can be found, and using the data of these similar patients to train the model again can make the model more in line with the actual situation of the target patient, thereby improving the accuracy of predicting the onset of diabetic foot in this patient; moreover, the temperature difference data of the similar patient group may contain some unique and subtle features that have not been fully explored in the general model; retraining the basic prediction model can allow the model to learn these more detailed features related to the onset of diabetic foot and further improve the prediction accuracy.

[0086] Embodiment 2, please refer to Figure 2 As shown, the present application provides a method for constructing a diabetic foot onset prediction model based on deep learning, including the following steps:

[0087] Step S1, collect the foot thermal imaging information and basic physiological information of diabetic patients, and construct a patient foot reference data set; Step S1 includes the following sub-steps:

[0088] Step S101: Select key parts of the foot, denoted as R1 to Rn respectively, and record the actual positions of R1 to Rn on the foot; and set the first acquisition method and the second acquisition method.

[0089] Step S102: The first acquisition method includes: Let the diabetic patient maintain a quiet state, fully expose the foot of the diabetic patient, and after the foot temperature stabilizes, for any one of the key parts of the foot, use a thermal imaging device to directly face the key part of the foot to be measured for thermal imaging shooting to obtain a static thermal imaging image.

[0090] Step S103: The second acquisition method includes: Fully expose the foot of the diabetic patient, and after the diabetic patient walks normally for t0 minutes, for any one of the key parts of the foot, use a thermal imaging device to directly face the key part of the foot to be measured for thermal imaging shooting to obtain a dynamic thermal imaging image.

[0091] Step S104: For any diabetic patient, obtain the gender, age, height, weight, and BIM of the diabetic patient, and at a first time interval, use the first acquisition method and the second acquisition method to obtain the static thermal imaging images and corresponding dynamic thermal imaging images of R1 to Rn of the diabetic patient, and at the same time record the acquisition time and the number of acquisitions. Denote the static thermal imaging image and the dynamic thermal imaging image of the same acquisition as a static-dynamic thermal imaging group, and sort them according to the acquisition time, which is denoted as the foot thermal imaging information of the diabetic patient.

[0092] Step S105: While obtaining the static thermal imaging image of the diabetic patient each time, collect the body temperature and heart rate of the diabetic patient, sort them according to the acquisition time, and merge and store them with the gender, age, height, weight, and BIM of the diabetic patient, which is denoted as the basic physiological information of the diabetic patient.

[0093] Step S106: Obtain the foot thermal imaging information and basic physiological information of multiple diabetic patients, and classify them according to the corresponding patients, which is denoted as the patient foot reference data set.

[0094] Step S2: Perform the first temperature processing on the foot thermal imaging information of the diabetic patient to obtain the foot temperature difference change data; Step S2 includes the following sub-steps:

[0095] Step S201: Perform temperature correction extraction processing on any thermal imaging image in the foot thermal imaging information of any diabetic patient.

[0096] Step S202: Temperature correction extraction processing; Step S202 includes the following sub-steps:

[0097] Step S2021: For any thermal imaging image, denoted as the first foot thermal image, divide the thermal imaging area corresponding to the key parts of the foot in the first foot thermal image, and denote it as the key part thermal imaging area;

[0098] Step S2022: Obtain the maximum value and the minimum value of the temperature values corresponding to all pixel points in the key part thermal imaging area, and denote them as Tma and Tmi respectively in sequence; Set the interval size as b, and divide the interval into multiple sub-intervals with a size of b, denoted as basic temperature sub-intervals;

[0099] Step S2023: Denote any pixel point in the key part thermal imaging area as the first central pixel point; Set the size of the neighborhood range as a*a, and with the first central pixel point as the center, obtain a neighborhood pixel area with a size of a*a, denoted as the first central neighborhood;

[0100] Step S2024: Obtain the temperature values corresponding to all pixel points in the first central neighborhood, count the basic temperature sub-intervals to which all temperature values belong respectively, and then count the basic temperature sub-interval with the most occurrences, denoted as the majority temperature interval;

[0101] Step S2025: Determine whether the basic temperature sub-interval corresponding to the corresponding first central pixel point belongs to the majority temperature interval. If it does not belong, and the ratio of the number of majority temperature intervals to the total number of basic temperature sub-intervals corresponding to the first central neighborhood exceeds K0, then mark the first central pixel point as an abnormal pixel point, otherwise mark the first central pixel point as a normal pixel point; Repeat the marking for all pixel points in the key part thermal imaging area;

[0102] Step S2026: Remove all abnormal pixel points in the key part thermal imaging area after marking, and obtain the temperature values corresponding to all normal pixel points, then calculate the average temperature value, and mark it as the corresponding static temperature value or dynamic temperature value according to the static thermal imaging image or dynamic thermal imaging image corresponding to the key part thermal imaging area;

[0103] Step S203: Repeat to obtain all the static temperature values or dynamic temperature values in all the foot thermal imaging information, and calculate the difference between the static temperature value and the dynamic temperature value corresponding to the same acquisition times of the same foot key part according to the first temperature difference formula, denoted as the static-dynamic temperature difference. The first temperature difference formula is as follows: R0 = AR - BR, where R0 represents the static-dynamic temperature difference, AR represents the static temperature value, and BR represents the corresponding dynamic temperature value;

[0104] Step S204: Obtain all the corresponding static-dynamic temperature differences in all the foot thermal imaging information. For any foot thermal imaging information, classify all the corresponding static-dynamic temperature differences according to the key foot parts they belong to, and sort them according to the corresponding collection time and collection times, which is denoted as the foot temperature difference change information.

[0105] Step S205: Repeatedly obtain the foot temperature difference change information of all patients, which is denoted as the foot temperature difference change data.

[0106] Step S3: Perform feature marking processing on the foot temperature difference change data to obtain the foot temperature sample data, and construct a basic prediction model for diabetic foot. Step S3 includes the following sub-steps:

[0107] Step S301: According to the foot temperature difference change data of the same key foot part of the same diabetic patient and the corresponding basic physiological information, form a risk feature vector with the static-dynamic temperature difference, heart rate, and body temperature at the same collection time, which is denoted as M = {m1, m2, m3}, where m1, m2, and m3 represent the static-dynamic temperature difference, heart rate, and body temperature in sequence, and sort the risk feature vector according to the corresponding collection times, which is denoted as the risk feature data of the corresponding key foot part.

[0108] Step S302: Obtain the risk feature data of all key foot parts of all diabetic patients, and mark them according to whether the corresponding diabetic patients have diabetic foot, which is denoted as the foot temperature sample data.

[0109] Step S303: Based on constructing the original prediction model, the original prediction model includes an input layer, an LSTM layer, a fully connected layer, and an output layer. Set the number of neurons in the input layer as c1, the number of neurons in the LSTM layer as c2, the number of neurons in the fully connected layer as c3, and the number of neurons in the output layer as c4, and set the activation function of the output layer as the sigmoid activation function.

[0110] Step S304: Use the foot temperature sample data to train the original prediction model, and after completion, obtain the basic prediction model.

[0111] Step S4: Screen the first similar patient according to the basic physiological information of the diabetic patient, construct a personalized prediction model for diabetic foot based on the foot temperature sample data of the first similar patient, and perform the onset risk prediction of diabetic foot. Step S4 includes the following sub-steps:

[0112] Step S401: For the diabetic patient to be predicted, denoted as the patient to be measured, obtain the gender, age, height, weight, and BIM of the patient to be measured, and set the age similarity threshold as e1, the height similarity threshold as e2, the weight similarity threshold as e3, and the BIM similarity threshold as e4.

[0113] Step S402: Mark other diabetic patients of the same gender as the patient to be tested and satisfying that the age difference is less than or equal to e1, the height difference is less than or equal to e2, the weight difference is less than or equal to e3, and the BIM difference is less than or equal to e4 as the suspicious similar patients of the patient to be tested; Obtain all the suspicious similar patients of the patient to be tested in the patient foot reference dataset;

[0114] Step S403: Obtain the foot thermal imaging information and basic physiological information of the key parts of the foot of the patient to be tested at the first time interval; Record them as the to-be-tested thermal imaging information and to-be-tested physiological information in sequence, and calculate the average heart rate and average body temperature of the patient to be tested according to the to-be-tested physiological information;

[0115] Step S404: Obtain the average heart rate and average body temperature of all corresponding suspicious similar patients according to the patient foot reference dataset;

[0116] Step S405: Among all the suspicious similar patients of the patient to be tested, mark the suspicious similar patients who simultaneously satisfy that the difference from the average heart rate of the patient to be tested is less than f1 and the difference from the average body temperature of the patient to be tested is less than f2 as the first similar patients, where f1 is the set heart rate similarity threshold and f2 is the set body temperature similarity threshold;

[0117] Step S406: Obtain the foot temperature sample data corresponding to the first similar patients, record it as the personalized sample data, and use the personalized sample data to train the basic prediction model. After completion, obtain the diabetic foot personalized prediction model of the basic prediction model;

[0118] Step S407: Perform the first temperature processing on the to-be-tested thermal imaging information of the patient to be tested, and perform feature marking processing, and then input it into the diabetic foot personalized prediction model according to the corresponding key parts of the foot respectively to obtain the disease risk probability of the selected key parts of the foot.

[0119] Example 3, please refer to Figure 5 as shown in Figure 5The structural schematic diagram of an electronic device is exemplified. The electronic device may include: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus. The memory stores computer-readable instructions. The processor can call the instructions in the memory. When the computer-readable instructions are executed by the processor, the steps in the method for constructing a diabetes foot onset prediction model based on deep learning are run to achieve the following functions: collecting the foot thermal imaging information and basic physiological information of diabetic patients, and constructing a reference data set of the patients' feet; performing a first temperature processing on the foot thermal imaging information of diabetic patients to obtain foot temperature difference change data; performing a feature marking process according to the foot temperature difference change data to obtain foot temperature sample data, and constructing a basic diabetes foot prediction model; screening the first similar patients according to the basic physiological information of diabetic patients, and constructing a personalized diabetes foot prediction model according to the foot temperature sample data of the first similar patients, and performing a diabetes foot onset risk prediction.

[0120] In addition, when the logical instructions in the above-mentioned memory are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0121] Example 4, the present application also provides a computer-readable storage medium. The present application provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the method for constructing a diabetes foot onset prediction model based on deep learning as described above are run to achieve the following functions: collecting the foot thermal imaging information and basic physiological information of diabetic patients, and constructing a reference data set of the patients' feet; performing a first temperature processing on the foot thermal imaging information of diabetic patients to obtain foot temperature difference change data; performing a feature marking process according to the foot temperature difference change data to obtain foot temperature sample data, and constructing a basic diabetes foot prediction model; screening the first similar patients according to the basic physiological information of diabetic patients, and constructing a personalized diabetes foot prediction model according to the foot temperature sample data of the first similar patients, and performing a diabetes foot onset risk prediction.

[0122] Through the description of the above embodiments, the embodiments of the present invention may be provided as a method, a system or a computer program product. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the prior art, may be embodied in the form of a software product, and the computer software product may be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0123] In the embodiments provided in the present application, it should be understood that the disclosed system or method may be implemented in other ways. The above-described embodiments are merely illustrative. For example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation. For another example, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other may be through some communication interfaces. The indirect coupling or communication connection of systems, modules and units may be electrical, mechanical or other forms.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for constructing a prediction model for the onset of diabetic foot based on deep learning, characterized in that, It includes the following steps: Collect the foot thermal imaging information and basic physiological information of diabetic patients, and construct a reference dataset of the patients' feet; Perform the first temperature processing on the foot thermal imaging information of diabetic patients to obtain the foot temperature difference change data; Perform feature marking processing according to the foot temperature difference change data to obtain foot temperature sample data, and construct a basic prediction model for diabetic foot; Screen the first similar patients according to the basic physiological information of diabetic patients, construct a personalized prediction model for diabetic foot according to the foot temperature sample data of the first similar patients, and predict the onset risk of diabetic foot.

2. The method for constructing a diabetic foot onset prediction model based on deep learning according to claim 1, wherein Collect the foot thermal imaging information and basic physiological information of diabetic patients, and the construction of the reference dataset of the patients' feet includes the following sub-steps: Select the key parts of the foot, denoted as R1 to Rn respectively, and record the actual positions of R1 to Rn on the foot; and set the first acquisition method and the second acquisition method; The first acquisition method includes: keeping the diabetic patient in a quiet state, fully exposing the foot of the diabetic patient, and after the foot temperature is stable, for any one of the key parts of the foot, use a thermal imaging device to directly face the key part of the foot to be measured for thermal imaging shooting to obtain a static thermal imaging image; The second acquisition method includes: fully exposing the foot of the diabetic patient, and after the diabetic patient walks normally for t0 minutes, for any one of the key parts of the foot, use a thermal imaging device to directly face the key part of the foot to be measured for thermal imaging shooting to obtain a dynamic thermal imaging image.

3. The method for constructing a diabetic foot onset prediction model based on deep learning according to claim 2, wherein Collect the foot thermal imaging information and basic physiological information of diabetic patients, and the construction of the reference dataset of the patients' feet includes the following sub-steps: For any diabetic patient, obtain the gender, age, height, weight and BMI of the diabetic patient, and at the first time interval, use the first acquisition method and the second acquisition method to obtain the static thermal imaging images and corresponding dynamic thermal imaging images of R1 to Rn of the diabetic patient, and record the acquisition time and the number of acquisitions at the same time. Denote the static thermal imaging image and the dynamic thermal imaging image collected at the same time as a static-dynamic thermal imaging group, and sort them according to the acquisition time, which is denoted as the foot thermal imaging information of the diabetic patient. The first time interval is t1; Collect the body temperature and heart rate of the diabetic patient at the same time when obtaining the static thermal imaging image of the diabetic patient each time, sort them according to the acquisition time, and merge and store them with the gender, age, height, weight and BMI of the diabetic patient, which is denoted as the basic physiological information of the diabetic patient; Obtain the foot thermal imaging information and basic physiological information of multiple diabetic patients, and classify them according to the corresponding patients, which is denoted as the reference dataset of the patients' feet.

4. The method for constructing a diabetic foot onset prediction model based on deep learning according to claim 3, characterized in that, Perform the first temperature processing on the foot thermal imaging information of diabetic patients to obtain the foot temperature difference change data, which includes the following sub-steps: For any one of the thermal imaging images in the foot thermal imaging information of a diabetic patient, temperature correction and extraction processing are performed. The temperature correction and extraction processing include: for any one thermal imaging image, denoted as the first foot thermal imaging, dividing the thermal imaging area corresponding to the key parts of the foot in the first foot thermal imaging, denoted as the key part thermal imaging area; obtaining the maximum value and the minimum value of the temperature values corresponding to all the pixel points in the key part thermal imaging area, denoted as Tma and Tmi respectively in sequence; setting the interval size as b, and dividing the interval into multiple sub-intervals with a size of b, denoted as basic temperature sub-intervals; For any pixel point in the thermal imaging area of the key part, it is denoted as the first central pixel point; set the neighborhood range size to a*a, and take the first central pixel point as the center to obtain a neighborhood pixel area of a*a size, which is denoted as the first central neighborhood; Obtain the temperature values corresponding to all pixel points in the first central neighborhood, count the basic temperature sub-intervals to which all temperature values belong respectively, and then count the basic temperature sub-interval with the most occurrences, denoted as the majority temperature interval. Determine whether the basic temperature sub-interval corresponding to the corresponding first central pixel point belongs to the majority temperature interval. If it does not belong, and the ratio of the number of majority temperature intervals to the total number of basic temperature sub-intervals corresponding to the first central neighborhood exceeds K0, then mark the first central pixel point as an abnormal pixel point; otherwise, mark the first central pixel point as a normal pixel point. Repeat the marking for all pixel points in the thermal imaging area of the key part. Remove all abnormal pixel points in the thermal imaging area of the key part after marking is completed, obtain the temperature values corresponding to all normal pixel points, then calculate the average temperature value, and mark it as the corresponding static temperature value or dynamic temperature value according to the static thermal imaging image or dynamic thermal imaging image corresponding to the thermal imaging area of the key part.

5. The method for constructing a diabetic foot onset prediction model based on deep learning according to claim 4, wherein, The first temperature processing of the foot thermal imaging information of diabetic patients to obtain the foot temperature difference change data further includes the following sub-steps: Repeatedly obtain all the static temperature values or dynamic temperature values in all the foot thermal imaging information, and calculate the difference between the static temperature value and the dynamic temperature value corresponding to the same acquisition time of the same foot key part according to the first temperature difference formula, denoted as the static-dynamic temperature difference. The first temperature difference formula is as follows: R0 = AR - BR, where R0 represents the static-dynamic temperature difference, AR represents the static temperature value, and BR represents the corresponding dynamic temperature value. Obtain all the corresponding static-dynamic temperature differences in all the foot thermal imaging information. For any foot thermal imaging information, classify all the corresponding static-dynamic temperature differences according to the foot key parts to which they belong, and sort them according to the corresponding acquisition time and acquisition times, denoted as the foot temperature difference change information. Repeatedly obtain the foot temperature difference change information of all patients, denoted as the foot temperature difference change data.

6. The method for constructing a diabetic foot onset prediction model based on deep learning according to claim 5, wherein Perform feature marking processing according to the foot temperature difference change data to obtain the foot temperature sample data, and construct the basic prediction model for diabetic foot, which further includes the following sub-steps: According to the foot temperature difference change data and the corresponding basic physiological information of the same foot key part of the same diabetic patient, form a risk feature vector with the static-dynamic temperature difference, heart rate, and body temperature at the same acquisition time, denoted as M = {m1, m2, m3}, where m1, m2, and m3 represent the static-dynamic temperature difference, heart rate, and body temperature in sequence, and sort the risk feature vector according to the corresponding acquisition times, denoted as the risk feature data of the corresponding foot key part. Obtain the risk feature data of all foot key parts of all diabetic patients, and mark them according to whether the corresponding diabetic patients have diabetic foot, denoted as the foot temperature sample data.

7. The method for constructing a diabetic foot onset prediction model based on deep learning according to claim 6, wherein Perform feature marking processing according to the foot temperature difference change data to obtain the foot temperature sample data, and construct the basic prediction model for diabetic foot, which further includes the following sub-steps: Based on constructing an original prediction model, the original prediction model includes an input layer, an LSTM layer, a fully connected layer, and an output layer. Set the number of neurons in the input layer as c1, the number of neurons in the LSTM layer as c2, the number of neurons in the fully connected layer as c3, and the number of neurons in the output layer as c4. And set the activation function of the output layer as the sigmoid activation function. Use the foot temperature sample data to train the original prediction model, and after completion, obtain the basic prediction model.

8. The method for constructing a diabetic foot onset prediction model based on deep learning according to claim 7, wherein, Screening the first similar patients according to the basic physiological information of diabetic patients includes the following sub-steps: For the diabetic patient to be predicted, denoted as the patient to be measured, obtain the gender, age, height, weight, and BMI of the patient to be measured, and set the age similarity threshold as e1, the height similarity threshold as e2, the weight similarity threshold as e3, and the BMI similarity threshold as e4; Mark other diabetic patients who have the same gender as the patient to be measured and at the same time satisfy that the age difference is less than or equal to e1, the height difference is less than or equal to e2, the weight difference is less than or equal to e3, and the BMI difference is less than or equal to e4 as the suspected similar patients of the patient to be measured; Obtain all the suspected similar patients of the patients to be measured in the patient foot reference dataset; Obtain the foot thermal imaging information and basic physiological information of the key parts of the foot of the patient to be measured at the first time interval; Denote them as the to-be-measured thermal imaging information and to-be-measured physiological information in sequence. Calculate the average heart rate and average body temperature of the patient to be measured according to the to-be-measured physiological information; According to the patient foot reference dataset, obtain the average heart rate and average body temperature of all corresponding suspected similar patients; Among all the suspected similar patients of the patient to be measured, mark the suspected similar patients who simultaneously satisfy that the difference from the average heart rate of the patient to be measured is less than f1 and the difference from the average body temperature of the patient to be measured is less than f2 as the first similar patients, where f1 is the set heart rate similarity threshold and f2 is the set body temperature similarity threshold.

9. The method for constructing a diabetic foot onset prediction model based on deep learning according to claim 8, characterized in that, Construct a diabetic foot personalized prediction model according to the foot temperature sample data of the first similar patients and conduct diabetic foot onset risk prediction, including the following sub-steps: Obtain the foot temperature sample data corresponding to the first similar patients, denoted as personalized sample data. Use the personalized sample data to train the basic prediction model, and after completion, obtain the diabetic foot personalized prediction model of the basic prediction model; Conduct the first temperature processing on the to-be-measured thermal imaging information of the patient to be measured, and conduct feature marking processing. Then input it into the diabetic foot personalized prediction model according to the corresponding key parts of the foot, and obtain the onset risk probability of the selected key parts of the foot.

10. A system for constructing a diabetic foot onset prediction model based on deep learning, applicable to the method for constructing a diabetic foot onset prediction model based on deep learning according to any one of claims 1-9, characterized in that, It includes an information collection module, an information processing module, a basic construction module, and a risk prediction module; The information collection module is used to collect the foot thermal imaging information and basic physiological information of diabetic patients and construct a patient foot reference dataset; The information processing module is used to conduct the first temperature processing on the foot thermal imaging information of diabetic patients to obtain foot temperature difference change data; The basic construction module is used to conduct feature marking processing according to the foot temperature difference change data to obtain foot temperature sample data and construct a diabetic foot basic prediction model; The risk prediction module includes a strengthening unit and a prediction unit. The strengthening unit screens the first similar patients according to the basic physiological information of diabetic patients, and constructs a personalized prediction model for diabetic foot based on the foot temperature sample data of the first similar patients. The prediction unit is used to predict the onset risk of diabetic foot.

Images