A non-invasive intelligent analysis method for human body iron ion concentration

Abstract

This invention discloses a non-invasive intelligent analysis method for human iron ion concentration. The method includes the following steps: First, dynamic sequence spectral data of the front and back of the finger joints and the corresponding iron ion concentration values ​​are acquired. Then, a sequence multi-angle learning module and a concentration analysis module are used to train an intelligent iron ion concentration analysis network to explore the potential relationship between sequence spectral data from different perspectives and iron ion concentration. Finally, based on the trained intelligent iron ion concentration analysis network, a non-invasive analysis of the iron ion concentration in the human body is performed using the finger joint spectral data. This method can achieve intelligent analysis of human iron ion concentration non-invasively and quickly, filling the gap in existing technologies that cannot use non-invasive methods to detect human iron ion concentration in real time.

Description

Technical Field

[0001] This invention relates to the fields of non-invasive detection and artificial intelligence, specifically to a non-invasive intelligent analysis method for human iron ion concentration. Background Technology

[0002] Iron, as one of the essential trace elements for human metabolism, is closely related to human health and disease. Abnormal iron levels in the body can cause many diseases, such as iron deficiency leading to weakened immune system function, infertility, and anemia; iron overload can cause various liver diseases, diabetes, and cardiovascular diseases. Currently, methods for detecting iron ion levels are quite mature, but they generally require invasive sampling of the patient, increasing the risk of infection and reducing the patient's medical experience. Therefore, non-invasive quantitative analysis of iron levels in the human body is of great significance for the diagnosis, treatment, and prevention of diseases.

[0003] According to medical definitions, the normal human body contains 3-5g of iron, with hemoglobin accounting for the largest proportion, approximately 70%. Iron is a crucial component in the production of hemoglobin, which forms red blood cells. Clinically, low hemoglobin density (LHD) in red blood cells is tested as a reference for diagnosing iron deficiency, demonstrating a direct correlation between iron levels and hemoglobin. Currently, based on the absorption characteristics of blood components to different wavelengths of light, non-invasive analysis and applications of hemoglobin have been achieved, providing a feasible approach for non-invasive prediction of human iron levels.

[0004] Therefore, this invention proposes a non-invasive intelligent analysis method for human iron ion concentration. By utilizing artificial intelligence technology, it fits the intrinsic relationship between human iron content and absorption spectrum, and completes the intelligent analysis of human iron ion concentration in a non-invasive and rapid manner, filling the gap in the existing technology that cannot use non-invasive methods to detect human iron ion concentration in real time. Summary of the Invention

[0005] To address the aforementioned issues in existing technologies, this paper proposes using artificial intelligence to fit the intrinsic relationship between human iron content and absorption spectrum, enabling non-invasive and rapid intelligent analysis of human iron ion concentration, thus filling the gap in existing technologies that cannot perform real-time detection of human iron ion concentration using non-invasive methods.

[0006] This invention proposes a non-invasive intelligent method for analyzing the concentration of iron ions in the human body:

[0007] Step S01: Using a non-invasive data acquisition method, two different dynamic sequence spectral data of the front and back of the finger joint are obtained; the corresponding iron ion concentration value is determined by an invasive blood sampling method.

[0008] Step S02: Train the intelligent analysis network for iron ion concentration based on the data obtained in step S01;

[0009] Step S03: Based on the intelligent analysis network for iron ion concentration trained in step S02, the network can perform non-invasive analysis of iron ion concentration in the human body using data from the front and back of the finger joints.

[0010] Further: In step S01, a non-invasive technique is used to acquire dynamic sequence data. By utilizing the characteristics of the absorption spectrum of hemoglobin in blood, an optical sensor is used to irradiate the finger joint area, the absorption spectrum of the reflected light is measured, and the sequence spectral data over a period of time is saved.

[0011] Each sample in the dataset consists of three parts: the frontal knuckle spectral dataset V f Reverse knuckle spectral dataset V b And the iron ion concentration label Y:

[0012]

[0013] In the formula, n represents the sample size, X f and X b The included sequence spectral data are:

[0014]

[0015] In the formula, t represents the length of the sequence spectral data.

[0016] Furthermore: In step S02, the intelligent iron ion concentration analysis network consists of a sequence multi-angle learning module and a concentration analysis module. The sequence multi-angle learning module is used to explore the potential connections between sequence spectral data from different perspectives, and the concentration analysis module is used to explore the relationship between the output of the sequence multi-angle learning module and the iron ion concentration. The intelligent iron ion concentration analysis network is an end-to-end structure, and the overall loss function is:

[0017]

[0018] in, For multi-angle learning loss of the sequence, This represents the loss in concentration analysis.

[0019] Furthermore, the training process of the aforementioned intelligent iron ion concentration analysis network includes the following steps:

[0020] Step A10, obtain the frontal knuckle spectral dataset V f and the reverse knuckle spectral dataset V b As input to the sequence multi-perspective learning module, the sequence multi-perspective learning module explores the relationship between data from different perspectives by learning positive and negative sample pairs;

[0021] Step A20: Use the output representation vector z of the sequence multi-angle learning module and the corresponding iron ion concentration label as input to the concentration analysis module to train the concentration analysis module;

[0022] Step A30: Update the parameters of the time series multi-angle learning module and the concentration analysis module by backpropagation in the direction where the loss function value decreases, and select the weight with the highest analysis accuracy as the optimal weight.

[0023] Furthermore, the learning process of the aforementioned sequence multi-angle learning module is as follows:

[0024] Frontal knuckle spectral dataset V f and the reverse knuckle spectral dataset V b The included samples are denoted as n is the number of samples, and the number of positive sample pairs is... Negative sample pairs Train a discriminant function that can accurately find positive sample pairs from a set containing one positive sample pair and k (k < n) negative samples. The discriminant is determined by calculating the distance between the sample pairs; the smaller the distance, the higher the score.

[0025] Furthermore, the discriminant function is implemented using two encoders. and For Vf and V b Perform feature transformation to obtain the corresponding feature representation, denoted as . and Then randomly fix V f The i-th sample in V, then enumerate V b For k samples, the objective loss function is as follows:

[0026]

[0027] Where τ is a hyperparameter used for dynamic adjustment range, θ f and θ b encoders and The parameter c(·) is the distance metric function;

[0028] In formula (4), the loss It is a fixed view V f The middle element is regarded as the anchor enumeration V b Obtained from the middle element. Symmetrically, the anchor point is set at V. b Enumeration V from the middle f middle element to obtain The comparison losses are then summed together as the sequence multi-angle learning module loss:

[0029]

[0030] After the sequence multi-angle learning stage, the feature representation z f and z b The concatenation yields a representation vector z, which serves as the input for the next stage.

[0031] Furthermore: the concentration analysis module is used to explore the relationship between the output of the sequence multi-angle learning module and the iron ion concentration. This module takes the representation vector z and the iron ion concentration label as input, and its loss function is:

[0032]

[0033] In the formula, Y represents the actual iron ion concentration, f p The vector z represents the mapping relationship between the vector z and the iron ion concentration, and p is a parameter for concentration analysis.

[0034] The beneficial effects of this invention are as follows: The non-invasive intelligent analysis method for human iron ion concentration in this invention acquires spectral data of the front and back knuckle sequences through a non-invasive acquisition method. The acquired dataset is used to train an intelligent analysis network for iron ion concentration. The multi-angle learning module of this network can explore the potential relationship between spectral data from different perspectives, and the concentration analysis module explores the relationship between the output of the multi-angle learning module and the iron ion concentration. After training, the intelligent analysis network for iron ion concentration can perform non-invasive analysis of the iron ion concentration in the human body based on the front and back knuckle data. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the method of the present invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings. The description in this part is only exemplary and explanatory, and should not be used to limit the scope of protection of the present invention in any way.

[0037] The core idea of ​​this invention is to design a suitable algorithm based on spectral sequence data of finger joints from different perspectives to mine the relationship between the data from different perspectives, maximize the effective information contained in the data, and then explore the mapping relationship between spectral sequence data and iron ion concentration, so as to achieve non-invasive analysis of iron ion concentration in the human body based on the front and back data of finger joints.

[0038] This invention provides a non-invasive intelligent method for analyzing the concentration of iron ions in the human body, such as... Figure 1 As shown, the method includes the following steps:

[0039] Step S01: Using a non-invasive data acquisition method, sequence spectral data from two different perspectives, the front and back of the finger joint, are obtained; the corresponding iron ion concentration is determined using an invasive blood sampling method.

[0040] Step S02: Train the intelligent analysis network for iron ion concentration based on the data obtained in step S01;

[0041] Step S03: Based on the intelligent analysis network for iron ion concentration trained in step S02, the network can perform non-invasive analysis of iron ion concentration in the human body using data from the front and back of the finger joints.

[0042] The non-invasive data acquisition method employs a non-invasive technique for dynamic sequence data acquisition. Utilizing the characteristics of the absorption spectrum of hemoglobin in blood, an optical sensor is used to irradiate the finger joint area, and the absorption spectrum of the reflected light is measured and the sequence spectral data over a period of time is stored. The corresponding iron ion concentration in the human body is determined using an invasive blood sampling method.

[0043] Each sample in the completed dataset consists of three parts: the frontal knuckle spectral dataset V f Reverse knuckle spectral dataset V b And the iron ion concentration label Y:

[0044]

[0045] In the formula, n represents the sample size, X f and X b The included sequence spectral data are:

[0046]

[0047] In the formula, t represents the length of the sequence spectral data;

[0048] Optionally, 70% of the data can be used as the training set, 10% as the validation set, and 20% as the test set.

[0049] The intelligent iron ion concentration analysis network in step S02 consists of a sequence multi-angle learning module and a concentration analysis module. The sequence multi-angle learning module is used to explore the potential relationships between sequence spectral data from different perspectives, and the concentration analysis module is used to explore the relationship between the output of the sequence multi-angle learning module and the iron ion concentration. The intelligent iron ion concentration analysis network is an end-to-end structure, and the overall loss function is:

[0050]

[0051] in, For the total loss, For multi-angle learning loss of the sequence, Loss due to concentration analysis;

[0052] The training process of the intelligent iron ion concentration analysis network includes the following steps:

[0053] Step A10, obtain the frontal knuckle spectral dataset V f and the reverse knuckle spectral dataset V b As input to the sequence multi-perspective learning module, the sequence multi-perspective learning module explores the potential connections between data from different perspectives by learning positive and negative sample pairs, maximizing the effective information contained in the data;

[0054] Step A20: Use the output representation vector z of the sequence multi-angle learning module and the corresponding iron ion concentration label as input to the concentration analysis module to train the concentration analysis module;

[0055] Step A30: Update the parameters of the time series multi-angle learning module and the concentration analysis module by backpropagation in the direction where the loss function value decreases, and select the weight with the highest analysis accuracy as the optimal weight;

[0056] Optionally, every A iterations, an accuracy test of the concentration analysis is performed on the validation set, and weights with an accuracy greater than a set threshold are retained. After the iterative training is completed, the weight with the highest analysis accuracy is selected as the optimal weight.

[0057] The goal of the sequence multi-angle learning module is to learn an embedding mapping that can establish a connection between data from different perspectives and maximize the interactive information under the same iron ion content. The input of this module is sequence spectral data.

[0058] The specific learning process of the sequence multi-angle learning module is as follows:

[0059] Frontal knuckle spectral dataset V f and the reverse knuckle spectral dataset V b The included samples are denoted as n is the number of samples, and the number of positive sample pairs is... Negative sample pairs Train a discriminant function so that it can accurately find positive sample pairs from a set containing 1 positive sample pair and k (k < n) negative samples. The discrimination is done by calculating the distance between the sample pairs, and the smaller the distance, the higher the score.

[0060] The discriminant function is implemented using two encoders. and Perform feature transformation on Vf and Vb to obtain the corresponding feature representations, denoted as . and Then randomly fix V fThe i-th sample in V, then enumerate V b For k samples, the objective loss function is as follows:

[0061]

[0062] Where τ is a hyperparameter used for dynamic adjustment range, θ f and θ b encoders and The parameter c(·) is the distance metric function;

[0063] Optionally, c(·) can use cosine similarity as a distance metric, as shown in the following formula:

[0064]

[0065] In formula (4), the loss It is a fixed view V f The middle element is regarded as the anchor enumeration V b Obtained from the middle element. Symmetrically, the anchor point is set at V. b Enumeration V from the middle f middle element to obtain The comparison losses are then summed together as the sequence multi-angle learning module loss:

[0066]

[0067] After the sequence multi-angle learning stage, the feature representation z f and z b The concatenation yields a representation vector z, which serves as the input for the next stage.

[0068] The concentration analysis module is used to explore the relationship between the output of the sequence multi-angle learning module and the iron ion concentration. This module takes the representation vector z and the iron ion concentration label as input, and the loss function of this module is:

[0069]

[0070] In the formula, Y represents the actual iron ion concentration, f p The vector z represents the mapping relationship between the vector z and the iron ion concentration, and p is a parameter for concentration analysis.

[0071] Optionally, a system utilizing an intelligent iron ion concentration analysis network can perform non-invasive analysis of iron ion concentration in the human body based on data from the front and back of the knuckles, and provide corresponding suggestions according to the human body's iron ion content standards, as follows:

[0072] Adult males: 11.0–30.0 μmol / L is normal;

[0073] Adult women: 9.0–27.0 μmol / L is normal;

[0074] Children: 9.0–22.0 μmol / L is normal;

[0075] For the elderly: 7.16–28.6 μmol / L is normal.

[0076] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0077] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only for the purpose of helping to understand the method and core ideas of the present invention. The above descriptions are only preferred embodiments of the present invention. It should be noted that due to the limitations of textual expression, there are objectively infinite specific structures. For those skilled in the art, several improvements, modifications, or changes can be made without departing from the principles of the present invention, and the above technical features can also be combined in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of the present invention.

Claims

1. A non-invasive intelligent method for analyzing the concentration of iron ions in the human body, characterized in that, Step S01: Obtain two different dynamic sequence spectral data of the front and back of the knuckle and the corresponding iron ion concentration values; Step S02: Train the intelligent analysis network for iron ion concentration based on the data obtained in step S01. The intelligent analysis network for iron ion concentration in step S02 consists of a sequence multi-angle learning module and a concentration analysis module. The sequence multi-angle learning module is used to explore the potential relationship between sequence spectral data from different perspectives, and the concentration analysis module is used to explore the relationship between the output of the sequence multi-angle learning module and the iron ion concentration. Step S03: Based on the intelligent analysis network for iron ion concentration trained in step S02, the network can perform non-invasive analysis of iron ion concentration in the human body using data from the front and back of the finger joints. The training process includes a step A10 of inputting a positive knuckle spectrum dataset and a negative knuckle spectrum dataset as input of a sequence multi-angle learning module, the sequence multi-angle learning module explores the connection between data between different perspectives through learning of positive and negative sample pairs; The learning process of the sequence multi-angle learning module is as follows: Frontal knuckle spectral dataset and reverse knuckle spectral dataset The included samples are denoted as , The sample size is and the number of positive sample pairs is . negative sample pairs Train a discriminant function so that it can distinguish between pairs containing one positive sample and... In a set of negative samples, the positive sample pairs are accurately found. The method of discrimination is to calculate the distance between the sample pairs. The smaller the distance, the higher the score. The discriminant function is implemented using two encoders. and right and Perform feature transformation to obtain the corresponding feature representation, denoted as . and Then randomly fixed The first in 1 sample, then enumerate In For each sample, the objective loss function is as follows: in These are hyperparameters, used to dynamically adjust the range. and encoders and The parameters, It is a distance metric function; In formula (4), the loss It is a fixed view The middle element is considered as an anchor enumeration. The elements obtained symmetrically set the anchor point at... Enumeration from the middle middle element to obtain The losses are compared, and then summed together as the loss for the sequence multi-angle learning module: After the sequence multi-angle learning stage, the feature representation is... and Concatenation yields the representation vector As input for the next stage.

2. The non-invasive intelligent analysis method for human iron ion concentration according to claim 1, characterized in that, In step S01, each sample in the dataset consists of three parts: the frontal knuckle spectral dataset. Reverse knuckle spectral dataset and iron ion concentration label : In the formula, Represents the number of samples. and The included sequence spectral data are: In the formula, This represents the length of the sequence spectral data.

3. The non-invasive intelligent analysis method for human iron ion concentration according to claim 1, characterized in that, The intelligent iron ion concentration analysis network is an end-to-end structure, and its overall loss function is: in, For multi-angle learning loss of the sequence, This represents the loss in concentration analysis.

4. The non-invasive intelligent analysis method for human iron ion concentration according to claim 3, the training process further includes the following steps: Step A20: Convert the output representation vector of the sequence multi-angle learning module. The corresponding iron ion concentration labels are used as inputs to train the concentration analysis module. Step A30: Update the parameters of the time series multi-angle learning module and the concentration analysis module by backpropagation in the direction where the loss function value decreases, and select the weight with the highest analysis accuracy as the optimal weight.

5. A non-invasive intelligent analysis method for human iron ion concentration according to claim 4, wherein the concentration analysis module is used to explore the relationship between the output of the sequence multi-angle learning module and the iron ion concentration, and the module uses a vector representation. With the iron ion concentration label as input, the loss function of this module is: In the formula, This represents the actual iron ion concentration. Represents a vector The mapping relationship between iron ion concentration and iron ion concentration. These are parameters for concentration analysis.

Images