Short-time blood glucose neural network prediction method of embedded vector decomposition model

Abstract

The invention provides a short-time blood glucose prediction method. The method comprises the following steps: acquiring a natural blood glucose sequence, a blood glucose increasing behavior sequence and a blood glucose reducing behavior sequence corresponding to the blood glucose of a patient, the sugar intake behavior of the patient and the active blood glucose reducing behavior of the patient; generating a corresponding natural blood glucose vector, a corresponding glycemic vector and a corresponding hypoglycemic vector according to the sequence; the blood glucose variation under natural blood regulation is obtained by averaging natural blood glucose vectors or performing neural network mapping, and the blood glucose variation caused by the blood glucose rising behaviors is obtained by performing convolution or neural network mapping on the blood glucose rising vectors and the blood glucose rising behavior sequence. Performing convolution or neural network mapping on the hypoglycemic vector and the hypoglycemic behavior sequence to obtain a hypoglycemic change amount caused by the hypoglycemic behavior; summing the blood glucose value at a certain moment with the blood glucose change amount under natural blood regulation and the blood glucose change amount caused by the blood glucose rising behavior and the blood glucose reducing behavior to obtain a blood glucose predicted value at a next moment.

Description

Technical Field

[0001] The present invention relates to the technical field of blood sugar prediction, and in particular to a short-term blood sugar prediction method based on an embedded vector decomposition model. Background Art

[0002] For diabetic patients, real-time blood sugar level is crucial for timely intervention and keeping blood sugar within a healthy range. Although continuous blood glucose monitors can provide real-time blood sugar data, they require long-term skin puncture, which not only increases the risk of infection, but may also have a negative impact on patients' psychology.

[0003] In view of this, there is a need to provide a method to achieve short-term blood sugar prediction without piercing the skin, thereby providing diabetic patients with a more convenient and safe means of blood sugar monitoring. Summary of the invention

[0004] The present invention provides a short-term blood glucose prediction method based on an embedded vector decomposition model, which is used to solve the problem in the prior art that current blood glucose monitoring requires long-term skin puncture, which increases the burden on patients, and realizes convenient and safe blood glucose monitoring.

[0005] The present invention provides a short-term blood sugar prediction method, comprising: Step 1, obtaining the natural blood sugar sequence, blood sugar raising behavior sequence and blood sugar lowering behavior sequence corresponding to the patient's blood sugar, the patient's sugar intake behavior and the patient's active blood sugar lowering behavior; Step 2, generating corresponding natural blood sugar vectors, blood sugar raising vectors and blood sugar lowering vectors according to the natural blood sugar sequence, the blood sugar raising behavior sequence and the blood sugar lowering behavior sequence; Step 3, obtaining the blood sugar change under the current natural blood regulation according to the natural blood sugar vector; obtaining the blood sugar change caused by the blood sugar raising behavior according to the blood sugar lowering vector and the blood sugar lowering behavior sequence; obtaining the blood sugar change caused by the blood sugar lowering behavior according to the blood sugar lowering vector and the blood sugar lowering behavior sequence; Step 4, summing the blood sugar change under natural blood regulation, the blood sugar change caused by blood sugar-raising behavior, and the blood sugar change caused by blood sugar-lowering behavior to obtain the blood sugar change; Step 5, sum the blood sugar change and the blood sugar value at a certain moment to obtain the blood sugar prediction value at the next moment.

[0006] As a specific embodiment of the present invention, step 2 further includes: Perform n-time function mapping on the blood glucose at time k to obtain an n-dimensional natural blood glucose vector.

[0007] As another specific embodiment of the present invention, step 2 further comprises: Step 21, input the blood glucose value at time k into the neural network; Step 22, performing embedded vector decomposition on the blood glucose value at time k through a neural network to obtain a natural blood glucose vector with a dimension of [1, n]; Step 23, using a neural network, the natural blood glucose vector is reduced in dimension to obtain the blood glucose change under natural blood regulation.

[0008] Wherein, step 2 further comprises, The blood sugar-raising behavior sequence is input into the neural network, and each value in the sequence is upgraded to n dimensions through the neural network to obtain a learnable blood sugar-raising vector.

[0009] The blood sugar lowering behavior sequence is input into the neural network, and each value in the sequence is upgraded to n dimensions through the neural network to obtain a learnable blood sugar lowering vector.

[0010] As a specific embodiment of the present invention, step 3 further includes: The blood sugar vectors at multiple moments are input into the recurrent neural network, and the blood sugar changes caused by the blood sugar behavior are obtained through model training.

[0011] The blood sugar lowering vectors at multiple moments are input into the recurrent neural network, and the blood sugar change caused by the blood sugar lowering behavior is obtained through model training.

[0012] As another specific embodiment of the present invention, step 3 further comprises: The blood sugar rise vector is convolved with the blood sugar rise behavior sequence to obtain the blood sugar change caused by the blood sugar rise behavior.

[0013] The blood sugar lowering vector is convolved with the blood sugar lowering behavior sequence to obtain the blood sugar change caused by the blood sugar lowering behavior.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program implements any of the above-mentioned short-time blood glucose prediction methods when executed by a processor.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, any of the above-mentioned short-time blood glucose prediction methods is implemented.

[0016] Specifically, the present invention proposes a short-term blood glucose prediction method based on an embedded vector decomposition model. By acquiring personal blood glucose, blood glucose-raising behavior data and blood glucose-lowering behavior data, the above-mentioned neural network, recurrent neural network, learnable blood glucose-raising vector embedding, and learnable blood glucose-lowering vector embedding can be trained to obtain a blood glucose prediction model. Based on the above-mentioned blood glucose prediction model, blood glucose prediction for the corresponding individual can be completed.

[0017] The blood sugar prediction method and blood sugar prediction model proposed in the present invention comprehensively consider the individual's blood sugar changes, sugar intake behavior and active blood sugar lowering behavior, and can accurately evaluate short-term blood sugar changes, avoiding the infection risks and psychological burden brought to patients by traditional blood sugar prediction methods that pierce the skin. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0019] Figure 1 It is a schematic diagram of the short-term blood sugar response modeling method provided by the present invention.

[0020] Figure 2 It is a schematic diagram of blood sugar prediction using a recursive neural network based on blood sugar modeling provided by the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings 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.

[0022] For diabetic patients, real-time blood sugar level is crucial for timely intervention and keeping blood sugar within a healthy range. Although continuous blood glucose monitors can provide real-time blood sugar data, they require long-term skin puncture, which not only increases the risk of infection, but may also have a negative impact on patients' psychology.

[0023] In general, the method of the present invention mainly includes three parts: using embedded vector decomposition to realize short-term blood sugar response modeling, using recursive neural network based on blood sugar modeling to predict blood sugar, and performing model calculation for blood sugar of different individuals. Through a series of algorithm processing, the present invention can realize accurate prediction of blood sugar values ​​and provide a more convenient and safe blood sugar monitoring method for diabetic patients.

[0024] Figure 1 Schematic diagram of the short-term blood sugar response modeling method of the present invention. Figure 1As shown, the present invention provides a short-term blood glucose prediction method based on an embedded vector decomposition model, comprising: step 1, obtaining the natural blood glucose sequence, blood glucose-raising behavior sequence and blood glucose-lowering behavior sequence corresponding to the patient's blood glucose, the patient's sugar intake behavior and the patient's active blood glucose-lowering behavior; step 2, generating corresponding natural blood glucose vectors, blood glucose-raising vectors and blood glucose-lowering vectors according to the natural blood glucose sequence, blood glucose-raising behavior sequence and blood glucose-lowering behavior sequence; step 3, obtaining the blood glucose change under the current natural blood glucose regulation according to the natural blood glucose vector; obtaining the blood glucose change caused by the blood glucose-raising behavior according to the blood glucose-raising vector and the blood glucose-raising behavior sequence; obtaining the blood glucose change caused by the blood glucose-lowering behavior according to the blood glucose-lowering vector and the blood glucose-lowering behavior sequence; step 4, summing the blood glucose change under natural blood glucose regulation, the blood glucose change caused by the blood glucose-raising behavior and the blood glucose-lowering behavior to obtain the blood glucose change; step 5, summing the blood glucose change with the blood glucose value at a certain moment to obtain the blood glucose prediction value at the next moment.

[0025] Among them, the patient's blood sugar, the patient's sugar intake behavior, and the patient's hypoglycemic drug intake behavior in step 1 can be collected in the following ways: The patient's blood sugar, sugar intake, and hypoglycemic drug intake are recorded at a fixed frequency. The frequency only needs to ensure that the time intervals are roughly the same. There is no restriction on the specific time intervals, and it can be recorded once per hour, once per minute, etc.

[0026] After multiple recordings, the following content is obtained: Blood glucose series[ , , , , …, ], the sequence can be obtained by sampling the data of 24-hour blood glucose recorder; Glycemic behavior sequence , , , , …, ], this sequence represents the amount of sugar the patient has eaten. No food is recorded as 0. The amount of food eaten can be converted according to a custom standard, such as the amount of glucose in grams, or the sugar ratio of 100 grams of steamed bread. For example, [0,0,0,0,1.5,0,0,0,0], converted according to the standard of the sugar ratio of 100 grams of steamed bread, means that the patient has eaten the amount of sugar equivalent to 1.5 100 grams of steamed bread at the 5th moment.

[0027] Glucose-lowering behavior sequence , , , , …, ] indicates the patient's active glucose-lowering behavior (such as insulin injection behavior, glucose-lowering drug taking behavior). If no active glucose-lowering behavior is performed, it will be recorded as 0. The glucose-lowering amount caused by glucose-lowering behavior can also be converted according to custom standards.

[0028] It should be noted that the above-mentioned description of collecting patient blood sugar, patient sugar intake behavior, and patient hypoglycemic drug intake behavior is only a means of collecting basic data and does not constitute a limitation on the scope of protection of this application.

[0029] According to a specific embodiment of the present invention, step 2 further comprises: The natural blood glucose vector is obtained by mapping the blood glucose at time k through a series of functions, and the mapping function can be a linear function, a quadratic function, a composite function, etc. The blood glucose increase vector and the blood glucose decrease vector can form a blood glucose increase curve or a blood glucose decrease curve through expert design, experimental sampling, etc.

[0030] Figure 2 Schematic diagram of blood glucose prediction based on recursive neural network blood glucose modeling provided by the present invention, such as Figure 2 As shown, according to another specific embodiment of the present invention, step 2 further includes: The neural network is embedded in the natural blood glucose vector construction, and a dimension-raising network with a dimension of [1,n] is constructed to raise the blood glucose from 1 dimension to n dimension. At the same time, activation functions such as relu and sigmoid and a fully connected network with an optional number of layers can be added, so as to act as a natural blood glucose vector mapping function through a neural network; then, a dimension-reducing network with a dimension of [n,1] is constructed to reduce the natural blood glucose vector to 1 dimension, and the blood glucose change under natural blood regulation is obtained. The following is a detailed description of it in conjunction with the embodiment: a dimension-raising fully connected neural network with network nodes of [1,50,50,80] and a dimension-reducing fully connected neural network with nodes of [80,50,50,1] are constructed, and a relu activation function is connected in series after each node. After the model is trained using personal blood glucose data obtained from a blood glucose meter, the training error can be as low as 3%, and the average prediction error can be as low as 7%.

[0031] The blood sugar-raising behavior sequence is input into the neural network, and each value in the sequence is upgraded to n dimensions through the neural network to obtain a learnable blood sugar-raising vector.

[0032] The blood sugar lowering behavior sequence is input into the neural network, and each value in the sequence is upgraded to n dimensions through the neural network to obtain a learnable blood sugar lowering vector.

[0033] According to a specific embodiment of the present invention, step 3 further comprises: The blood sugar vectors at multiple moments are input into the recurrent neural network, and the blood sugar changes caused by the blood sugar behavior are obtained through model training; Input the hypoglycemic vectors at multiple moments into a recurrent neural network, and through model training, obtain the blood glucose change amount caused by hypoglycemic behavior.

[0034] According to another specific embodiment of the present invention, step 3 further includes: Convolve the blood glucose rising vector with the blood glucose rising behavior sequence to obtain the blood glucose change amount caused by blood glucose rising behavior; convolve the hypoglycemic vector with the hypoglycemic behavior sequence to obtain the blood glucose change amount caused by hypoglycemic behavior.

[0035] Specifically, the present invention proposes a short-term blood glucose prediction method based on an embedded vector decomposition model. By acquiring personal blood glucose, blood glucose rising behavior data, and hypoglycemic behavior data, the above-mentioned neural network and recurrent neural network can be trained, and the learnable blood glucose rising vector embedding and learnable hypoglycemic vector embedding can be obtained to obtain a blood glucose prediction model. Based on the above blood glucose prediction model, the blood glucose prediction of the corresponding individual can be completed.

[0036] The blood glucose prediction method and blood glucose prediction model proposed by the present invention comprehensively consider the individual's blood glucose fluctuation situation, sugar intake behavior, and active hypoglycemic behavior, avoiding one-sided views on blood glucose problems. This comprehensive consideration method enables the method to accurately evaluate the short-term blood glucose change amount. Compared with the traditional method of pricking the skin to obtain blood glucose for prediction, it has significant advantages, effectively avoiding the infection risk and resulting psychological burden brought to patients, and providing a safer and more reliable solution for the blood glucose monitoring task.

[0037] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a short-term blood glucose prediction method based on an embedded vector decomposition model provided by the above-mentioned various methods, including: step 1, acquire the natural blood glucose sequence, blood glucose rising behavior sequence, and hypoglycemic behavior sequence corresponding to the patient's blood glucose, the patient's sugar intake behavior, and the patient's active hypoglycemic behavior; step 2, generate corresponding natural blood glucose vectors, blood glucose rising vectors, and hypoglycemic vectors according to the natural blood glucose sequence, blood glucose rising behavior sequence, and hypoglycemic behavior sequence; step 3, obtain the blood glucose change amount under the current natural blood glucose regulation according to the natural blood glucose vector; obtain the blood glucose change amount caused by blood glucose rising behavior according to the blood glucose rising vector and the blood glucose rising behavior sequence; obtain the blood glucose change amount caused by hypoglycemic behavior according to the hypoglycemic vector and the hypoglycemic behavior sequence; step 4, sum up the blood glucose change amount under the natural blood glucose regulation, the blood glucose change amount caused by blood glucose rising behavior, and the blood glucose change amount caused by hypoglycemic behavior to obtain the blood glucose change amount; step 5, sum up the blood glucose change amount and the blood glucose value at a certain moment to obtain the blood glucose prediction value at the next moment.

[0038] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the processor executes a short-term blood glucose prediction method based on an embedded vector decomposition model provided by the above-mentioned methods, including: step 1, obtaining the natural blood glucose sequence, blood glucose-raising behavior sequence and blood glucose-lowering behavior sequence corresponding to the patient's blood glucose, the patient's sugar intake behavior and the patient's active blood glucose-lowering behavior; step 2, generating corresponding natural blood glucose vectors, blood glucose-raising vectors and blood glucose-lowering vectors according to the natural blood glucose sequence, blood glucose-raising behavior sequence and blood glucose-lowering behavior sequence; step 3, obtaining the blood glucose change under the current natural blood glucose regulation according to the natural blood glucose vector; obtaining the blood glucose change caused by the blood glucose-raising behavior according to the blood glucose-raising vector and the blood glucose-raising behavior sequence; obtaining the blood glucose change caused by the blood glucose-lowering behavior according to the blood glucose-lowering vector and the blood glucose-lowering behavior sequence; step 4, summing the blood glucose change under natural blood glucose regulation, the blood glucose change caused by the blood glucose-raising behavior and the blood glucose-lowering behavior to obtain the blood glucose change; step 5, summing the blood glucose change with the blood glucose value at a certain moment to obtain the blood glucose prediction value at the next moment.

[0039] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0040] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can 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.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A short-term blood sugar prediction method, characterized in that: include: Step 1, obtaining the natural blood sugar sequence, blood sugar raising behavior sequence and blood sugar lowering behavior sequence corresponding to the patient's blood sugar, the patient's sugar intake behavior and the patient's active blood sugar lowering behavior; Step 2, generating corresponding natural blood sugar vectors, blood sugar raising vectors and blood sugar lowering vectors according to the natural blood sugar sequence, blood sugar raising behavior sequence and blood sugar lowering behavior sequence; Step 3, obtaining the blood sugar change under the current natural blood regulation according to the natural blood sugar vector; obtaining the blood sugar change caused by the blood sugar raising behavior according to the blood sugar raising vector and the blood sugar raising behavior sequence; obtaining the blood sugar change caused by the blood sugar lowering behavior according to the blood sugar lowering vector and the blood sugar lowering behavior sequence; Step 4, summing the blood sugar change under natural blood regulation, the blood sugar change caused by blood sugar-raising behavior, and the blood sugar change caused by blood sugar-lowering behavior to obtain the blood sugar change; Step 5, sum the blood sugar change and the blood sugar value at a certain moment to obtain the blood sugar prediction value at the next moment.

2. The method according to claim 1, characterized in that Step 2 also includes: performing n-time function mapping on the blood glucose at time k to obtain an n-dimensional natural blood glucose vector.

3. The method according to claim 2, characterized in that Step 2 further comprises, Step 21, input the blood glucose value at time k into the neural network; Step 22, performing embedded vector decomposition on the blood glucose value at time k through the neural network to obtain a natural blood glucose vector with a dimension of [1, n]; Step 23, reducing the dimension of the natural blood glucose vector through the neural network to obtain the blood glucose change under natural blood regulation.

4. The method according to claim 3, characterized in that Step 2 further comprises, The blood sugar-raising behavior sequence is input into a neural network, and each value in the sequence is dimensionally upgraded to n dimensions by the neural network to obtain a learnable blood sugar-raising vector.

5. The method according to claim 3, characterized in that: Step 2 further comprises, The blood sugar lowering behavior sequence is input into a neural network, and each value in the sequence is upgraded to n dimensions by the neural network to obtain a learnable blood sugar lowering vector.

6. The method according to claim 4, characterized in that Step 3 further comprises, The blood sugar vectors at multiple moments are input into the recurrent neural network, and the blood sugar changes caused by the blood sugar behavior are obtained through model training.

7. The method according to claim 5, characterized in that Step 3 further comprises, The blood sugar lowering vectors at multiple moments are input into the recurrent neural network, and the blood sugar change caused by the blood sugar lowering behavior is obtained through model training.

8. The method according to claim 4, characterized in that Step 3 further comprises, The blood sugar increase vector is convolved with the blood sugar increase behavior sequence to obtain the blood sugar change caused by the blood sugar increase behavior.

9. The method according to claim 5, characterized in that Step 3 further comprises, The blood sugar lowering vector is convolved with the blood sugar lowering behavior sequence to obtain the blood sugar change caused by the blood sugar lowering behavior.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the short-term blood glucose prediction method based on the embedded vector decomposition model as described in any one of claims 1 to 9 is implemented.

Images