An artificial intelligence-based method and system for classifying diseases in children

Abstract

This invention discloses an artificial intelligence-based method and system for classifying pediatric diseases, belonging to the field of medical artificial intelligence technology. The method includes the following steps: Step S1, collecting spectral data of a child's palm skin under different physiological conditions three times non-invasively using a spectrometer; Step S2, preprocessing the three collected spectral data respectively; Step S3, extracting trace element features related to children's health from the preprocessed spectral data; Step S4, constructing feature vectors based on the three collected data; Step S5, training the feature vectors using an artificial intelligence model based on reinforcement learning; Step S6, classifying new pediatric spectral data using the trained artificial intelligence model, outputting disease categories and probability distributions; Step S7, performing risk assessment based on the classification results, and generating a risk assessment report. This invention can obtain richer physiological information and improve the accuracy of disease classification.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence technology, specifically to a method and system for classifying childhood diseases based on artificial intelligence. Background Technology

[0002] Childhood illnesses, such as tumors, intellectual disability, and ADHD, have a profound impact on a child's growth and development. Early detection and intervention are crucial for improving prognosis. Currently, screening and diagnosis of these diseases mainly rely on the clinical experience of physicians, imaging examinations (such as CT and MRI), and laboratory biochemical tests. While these methods are effective, they have some limitations: imaging examinations may involve radiation risks and are expensive; laboratory tests are mostly invasive procedures, which are not easily accepted by children; and the diagnostic process heavily relies on the physician's subjective judgment, which may vary between different medical institutions.

[0003] In recent years, non-invasive detection technologies, such as spectral analysis, have shown great potential in the field of medical diagnostics due to their safety and speed. The content of trace elements (such as iron, zinc, and calcium) in human skin tissue is closely related to various physiological functions and disease states. By non-invasively collecting skin spectral data using a spectrometer, the levels of these trace elements can be indirectly reflected.

[0004] However, existing disease identification methods based on spectral analysis are mostly limited to data collection and analysis under single static conditions. The physiological state of the human body is dynamic, and single static data cannot fully reflect the functional changes of tissues under complex pathological conditions, leading to insufficient feature extraction and limiting the accuracy and robustness of classification models. Therefore, there is an urgent need for a non-invasive detection method that can acquire richer physiological information and improve the accuracy of disease classification.

[0005] To address these issues, those skilled in the art have provided an artificial intelligence-based method and system for classifying childhood diseases, in order to resolve the problems mentioned in the background section. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based method for classifying childhood diseases, comprising the following steps: Step S1: Collect spectral data of the child's palm skin in three non-invasive ways under different physiological conditions using a spectrometer. The first collection is performed when the child's palm is naturally straight and the five fingers are together in a relaxed state. The second collection is performed when the child's palm is forcefully straight and the five fingers are spread out to the maximum extent. The third collection is performed when the child's hand is gripping the grip strength meter and continuously exerting force until the preset grip strength threshold is reached. Step S2: Preprocess the spectral data acquired in the three acquisitions, including noise reduction, baseline correction and normalization. Step S3: Extract trace element features related to children's health from the preprocessed spectral data, wherein the trace elements include iron, zinc, calcium, magnesium, copper and selenium; Step S4: Construct a feature vector based on the data from the three acquisitions, including the static features corresponding to each acquisition and the dynamic response features between different acquisition states; Step S5: Train the feature vector using an artificial intelligence model based on reinforcement learning, using a historical dataset with disease labels. Step S6: Use the trained artificial intelligence model to classify the new pediatric spectral data and output the disease categories and probability distributions; Step S7: Conduct a risk assessment based on the classification results and generate a risk assessment report.

[0007] As a further aspect of the present invention: the spectral data collected in the three acquisitions cover the visible to near-infrared bands, with a wavelength range of 400 nanometers to 2500 nanometers; during acquisition, a standardized protocol is used to control environmental conditions, including temperature, humidity, and light intensity; the preset grip strength threshold of the grip strength meter is dynamically adjusted according to the child's age, gender, and weight percentile to ensure the standardization and comparability of the load state acquisition.

[0008] As a further aspect of the present invention: the denoising process in the preprocessing step adopts the moving average filtering method or the wavelet transform denoising algorithm; the baseline correction adopts the polynomial fitting algorithm with a goodness of fit of not less than 0.95; the normalization process adopts the min-max normalization method or the z-score standardization method to unify the data scale.

[0009] As a further aspect of the present invention: the extraction of trace element characteristics is achieved by analyzing the absorption peak, reflectance curve and spectral integral area at a specific wavelength; the dynamic response characteristics include the difference in spectral intensity, ratio, differential spectrum or integral area change rate between any two data acquisitions.

[0010] As a further aspect of the present invention: the reinforcement learning-based artificial intelligence model adopts a deep reinforcement learning network architecture, including an actor-critic architecture or a deep Q-network; during training, the feature vector is used as the state input, the disease classification result is used as the action space, and the model is guided to optimize through a reward function.

[0011] As a further aspect of the present invention: the training process in step S5 also includes an experience replay mechanism and a target network update strategy to improve training stability and convergence speed; the reward function is set according to the classification accuracy, with positive rewards for correct classification and negative rewards for incorrect classification.

[0012] As a further aspect of the present invention: the disease categories output in step S6 include health, tumor, intellectual disability and ADHD; wherein tumor is further subdivided into malignant tumor and benign tumor, and intellectual disability and ADHD are classified into mild, moderate and severe according to their severity.

[0013] As a further aspect of the present invention: the risk assessment step assesses the probability of a child contracting the disease by calculating a risk score, which is calculated based on classification probability, disease weighting factor, and clinical adjustment item; the risk assessment report includes risk level, recommended measures, and confidence index.

[0014] As a further aspect of the present invention, it also includes model validation and optimization steps: evaluating model performance using an independent test dataset, with evaluation metrics including accuracy, recall, F1 score, and AUC value; and optimizing model parameters through cross-validation and hyperparameter tuning.

[0015] This application also discloses an artificial intelligence-based classification system for childhood diseases, employing an artificial intelligence-based classification method for childhood diseases, including: The data acquisition module uses a spectrometer to collect spectral data of children's palm skin in three non-invasive ways under different physiological conditions. The first acquisition is carried out in a relaxed state with the child's palm naturally straight and the five fingers together. The second acquisition is carried out in a stretched state with the child's palm forcefully straight and the five fingers spread out to the maximum extent. The third acquisition is carried out when the child's hand holds the grip strength meter and continues to exert force until the preset grip strength threshold is reached. The preprocessing module performs preprocessing on the spectral data acquired in the three acquisitions, including noise reduction, baseline correction, and normalization. The feature extraction module extracts trace element features related to children's health from the preprocessed spectral data. The trace elements include iron, zinc, calcium, magnesium, copper, and selenium. The feature vector construction module constructs feature vectors based on data collected from three collections, including static features corresponding to each collection and dynamic response features between different collection states. The model training module uses a reinforcement learning-based artificial intelligence model to train the feature vectors, and the training uses a historical dataset with disease labels. The data classification module uses a trained artificial intelligence model to classify new pediatric spectral data and output disease categories and probability distributions. The risk assessment module performs risk assessments based on the classification results and generates risk assessment reports.

[0016] The beneficial effects of this invention are reflected in: 1. Richer information dimensions: By collecting spectral data under three different physiological states (relaxation, stretching, and load), it is possible to capture the dynamic response information of tissues under static, micro-change, and high load conditions, which greatly enriches the feature dimensions and provides more sufficient evidence for disease identification.

[0017] 2. Higher classification accuracy: The extracted dynamic response features are more sensitive to physiological functional abnormalities in pathological states. Combined with a reinforcement learning-based AI model, it can autonomously learn the optimal classification strategy from high-dimensional complex features, significantly improving the accuracy, recall, and generalization ability of disease classification.

[0018] 3. Non-invasive, safe, and easy to promote: The entire testing process is completely non-invasive, avoiding the pain and risks associated with radiation and blood collection, making it particularly suitable for early screening and regular monitoring of children. The method is highly standardized, facilitating its application in medical institutions at different levels.

[0019] 4. Assisting in diagnostic decision-making: The final output includes not only simple classification results, but also probability distributions and detailed risk assessment reports, providing clinicians with quantitative decision support and helping to achieve early warning and personalized intervention for diseases. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0021] Figure 1 A flowchart of an artificial intelligence-based method for classifying childhood diseases; Figure 2 This is a structural diagram of an artificial intelligence-based classification system for childhood diseases. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] As mentioned in the background section of this application, research has found that existing disease identification methods based on spectral analysis are mostly limited to data collection and analysis under single static conditions. The physiological state of the human body is dynamic and constantly changing; single static data cannot fully reflect the functional changes of tissues under complex pathological conditions, leading to insufficient feature extraction and limiting the accuracy and robustness of classification models, thus exhibiting certain shortcomings.

[0024] To address the aforementioned shortcomings, this application discloses an artificial intelligence-based method and system for classifying childhood diseases, which can acquire richer physiological information and improve the accuracy of disease classification.

[0025] The following will describe in detail, with reference to the accompanying drawings, how the solution of this application solves the above-mentioned technical problems.

[0026] Please see Figure 1 In this embodiment of the invention, an artificial intelligence-based method for classifying childhood diseases includes the following steps: Step S1: Collecting spectral data of a child's palm skin in three non-invasive steps using a spectrometer under different physiological states. The first collection is performed when the child's palm is naturally extended and the fingers are together in a relaxed state; the second collection is performed when the child's palm is forcefully extended and the fingers are spread out to the maximum extent; and the third collection is performed when the child's hand is gripping a grip strength meter and continuously exerting force until a preset grip strength threshold is reached; Step S2: Preprocessing the three collected spectral data, including noise reduction, baseline correction, and normalization; Step S3: From... The preprocessed spectral data is used to extract trace element features related to children's health, including iron, zinc, calcium, magnesium, copper, and selenium. Step S4 involves constructing a feature vector based on data from three acquisitions, including static features corresponding to each acquisition and dynamic response features between different acquisition states. Step S5 involves training the feature vector using a reinforcement learning-based artificial intelligence model, using a historical dataset labeled with diseases. Step S6 involves classifying new pediatric spectral data using the trained artificial intelligence model, outputting disease categories and probability distributions. Step S7 involves conducting a risk assessment based on the classification results and generating a risk assessment report. This application establishes the basic framework for a pediatric disease classification method based on multi-state spectral acquisition and artificial intelligence analysis. By acquiring spectral data under three different physiological states, combined with preprocessing, feature extraction, and reinforcement learning model training, it achieves automatic classification and risk assessment of pediatric diseases, providing a basic scope of protection for subsequent claims.

[0027] In this embodiment, the spectral data collected three times covered the visible to near-infrared band, with a wavelength range of 400 nm to 2500 nm. Standardized protocols were used to control environmental conditions during data acquisition, including temperature, humidity, and light intensity. The preset grip strength threshold of the hand dynamometer was dynamically adjusted based on the child's age, gender, and weight percentile to ensure standardization and comparability of data collected under different load conditions. This setup refines the acquisition parameters and environmental control requirements for spectral data. By standardizing acquisition conditions and dynamically adjusting the grip strength threshold, it ensures the consistency and comparability of data under different acquisition conditions, providing a high-quality data foundation for subsequent feature extraction and model training.

[0028] In this embodiment, the denoising process in the preprocessing step employs a moving average filtering method or a wavelet transform denoising algorithm; baseline correction uses a polynomial fitting algorithm with a goodness of fit of no less than 0.95; and normalization uses a min-max normalization method or a z-score normalization method to ensure uniform data scale. This setting specifically specifies the technical solutions and performance indicators used in each stage of the preprocessing process. By optimizing the denoising, baseline correction, and normalization methods, the quality and consistency of the spectral data are effectively improved, creating conditions for accurate feature extraction.

[0029] In this embodiment, the extraction of trace element characteristics is achieved by analyzing absorption peaks, reflectance curves, and spectral integral areas at specific wavelengths; dynamic response characteristics include the difference, ratio, differential spectrum, or rate of change of integral area between any two data acquisitions. More specifically, 1. Analysis based on absorption peaks at specific wavelengths: Each trace element, due to its unique atomic structure and electronic energy levels, selectively absorbs light energy at specific wavelengths when irradiated. This characteristic results in characteristic "absorption peaks" on the spectrum. In implementation, firstly, based on a known spectral database, one or more characteristic absorption wavelengths are determined for each trace element to be measured (iron, zinc, calcium, magnesium, copper, selenium). For example, iron may have a significant absorption peak at approximately 500 nanometers. Subsequently, the algorithm accurately locates these specific wavelength points in the preprocessed spectral data and reads their corresponding spectral intensity values ​​(i.e., absorption peaks). This intensity value directly reflects the degree of absorption of the element by the incident light; the stronger the absorption, the lower (in the reflectance spectrum) or the higher (in the absorption spectrum) the peak value, thus indirectly indicating the relative concentration of the element. By extracting the intensity values ​​of these key points, a set of the most direct elemental characteristics can be obtained. 2. Analysis based on reflectance curve morphology: Peak values ​​at a single wavelength may be susceptible to noise interference, therefore, more robust features need to be extracted from a more macroscopic curve morphology. The reflectance curve depicts the skin's reflectivity at different wavelengths. Variations in the content of trace elements affect the overall optical properties of the tissue in which they reside, causing a systematic change in the morphology of the reflectance curve within a specific wavelength range (not just a single point). In implementation, the algorithm focuses on a continuous wavelength range near the characteristic absorption peak of each element (e.g., the analysis range for iron might be 480 nm to 520 nm) and performs in-depth analysis of the reflectance curve within this range. This includes, but is not limited to: calculating the slope change of the curve within this range, identifying the existence of multiple consecutive small peaks or troughs (i.e., second derivative analysis), and evaluating the overall width and symmetry of the curve. These morphological parameters reflect the chemical environment of the element and its interaction with other tissues, providing more discriminative information beyond a single absorption peak. 3. Spectral Integral Area Analysis: To obtain information on the overall contribution of an element over a wider spectral range, integral area analysis is employed. The physical meaning of this method is to calculate the area enclosed by the spectral curve and the baseline within a specific wavelength range. In implementation, for each trace element, a relatively wide wavelength range associated with its absorption band is defined. The algorithm calculates the integral area below the reflectance curve (or a transformed curve, such as an absorbance curve) within this predefined range. For example, the spectral integral area of ​​zinc in the 580 nm to 620 nm wavelength range can be calculated.This area value is a cumulative measure, reflecting the total absorption or reflection intensity of the element within that spectral range. It is crucial for overcoming random fluctuations and enhancing the stability and robustness of the feature. In summary, the realization of this trace element feature is a multi-faceted and multi-layered information extraction process: first, the core feature points of the element are identified through absorption peaks; then, local detail changes are analyzed through the reflectance curve morphology; and finally, the macroscopic cumulative effect is obtained through the spectral integral area. These three types of features complement and verify each other, collectively forming a numerical feature vector that can comprehensively and stably characterize the distribution and state of various trace elements in the palm skin, laying a solid data foundation for subsequent high-precision disease classification using artificial intelligence models. This setup clarifies the specific implementation method of trace element feature extraction and the calculation method of dynamic response features. Through comparative analysis of multi-state data, richer information on physiological state changes can be captured, enhancing the feature's representational ability.

[0030] In this embodiment, the reinforcement learning-based artificial intelligence model employs a deep reinforcement learning network architecture, including an actor-critic architecture or a deep Q-network. During training, feature vectors are used as state inputs, and disease classification results are used as the action space. The model is guided to optimize through a reward function. This setup limits the specific architecture and training mechanism of the artificial intelligence model. By using a deep reinforcement learning network combined with reward function optimization, the model can continuously improve its classification strategy through interaction with the environment, thereby enhancing the accuracy of disease identification.

[0031] In this embodiment, the training process in step S5 also includes an experience replay mechanism and a target network update strategy to improve training stability and convergence speed. The reward function is set according to the classification accuracy, with positive rewards for correct classification and negative rewards for incorrect classification. This setting further refines the technical details of the training process, improves training stability through the experience replay mechanism and target network update strategy, and promotes the reliability of the model output by combining the confidence reward mechanism.

[0032] In this embodiment, the disease categories output in step S6 include health, tumor, intellectual disability, and ADHD; tumor is further subdivided into malignant tumors and benign tumors, and intellectual disability and ADHD are classified into mild, moderate, and severe according to their severity. This setting refines the specific categories and grading standards for disease classification, making the classification results more clinically valuable and providing a clear basis for subsequent risk assessment and medical intervention.

[0033] In this embodiment, the risk assessment step evaluates the probability of a child's illness by calculating a risk score, which is derived based on classification probability, disease weighting factors, and clinical adjustment items. The risk assessment report includes risk level, recommended actions, and confidence level indicators. This setup specifies the detailed calculation method and report content for risk assessment, providing quantitative decision support for clinical diagnosis and health management through multi-factor weighted assessment and grading. The calculation of this risk score is a multi-factor, weighted fusion decision-making process, the core of which lies in combining the classification output of the artificial intelligence model with prior clinical knowledge to generate a quantitative and interpretable risk indicator. The specific calculation process can be described through the following steps and components: Step 1. Determine the base probability: The calculation process begins with the classification probability output by the artificial intelligence model. For each disease category to be assessed (e.g., the risk of developing malignant tumors), the model outputs a probability value P between 0 and 1, which represents the initial probability that the child has this type of disease based on the current spectral characteristics. The higher the P value, the greater the risk perceived by the model. Step 2. Apply disease weighting factors: Different diseases have different clinical severity and urgency. To this end, the system pre-determines a disease weighting factor W for each disease. This factor is set based on large-scale epidemiological data, clinical guidelines, and expert consensus. For example, the weighting factor W for malignant tumors... cancer The weighting factor W for mild ADHD will be set very high (e.g., 10.0), while the weighting factor W for mild ADHD will be set very high. adhd-mild The risk score is relatively low (e.g., 1.5). This step is achieved by multiplying the base probability by the weighting factor (P×W), aiming to amplify the contribution of serious diseases in the risk score and ensure that high-priority clinical risks are significantly highlighted. Step 3. Introducing Clinical Adjustment Items: To provide personalized risk assessment, the system introduces a clinical adjustment item C. This is a comprehensive correction value calculated based on individualized clinical information such as age, gender, and family history mentioned in the application. For example, the algorithm may pre-set: for children with a family history of relevant diseases, add a positive adjustment value (e.g., +0.5); for children in a high-incidence age group for a certain disease, add another adjustment value. These adjustments are derived from medically confirmed risk factors, and they are incorporated into the total risk score either linearly or through a small adjustment model. Step 4. Summarizing and Calculating the Risk Score: The final risk score R is calculated by summing all the above components. Its basic calculation principle can be expressed as: Risk Score R = (Model Classification Probability P × Disease Weighting Factor W) + Clinical Adjustment Item C. For multi-disease classification scenarios, the system will independently calculate the risk score R for each disease. diseaseStep 5. Risk Level Mapping and Report Generation: The calculated raw risk score R is mapped to a preset risk level. For example, the system sets thresholds: R < 5 for "low risk", 5 ≤ R < 15 for "medium risk", and R ≥ 15 for "high risk". Ultimately, all this information—including the final risk score for each disease, the corresponding risk level, the probability and adjustment factors, and personalized recommendations—is integrated to generate a structured risk assessment report for medical professionals.

[0034] This embodiment also includes model validation and optimization steps: evaluating model performance using an independent test dataset, with evaluation metrics including accuracy, recall, F1 score, and AUC; and optimizing model parameters through cross-validation and hyperparameter tuning. This setup adds model validation and optimization steps, ensuring system reliability and generalization ability through rigorous performance evaluation and parameter tuning, providing quality assurance for practical applications.

[0035] This application also discloses an artificial intelligence-based classification system for childhood diseases, employing an AI-based classification method for childhood diseases, comprising: a data acquisition module, which non-invasively acquires spectral data of a child's palm skin under different physiological states three times using a spectrometer; the first acquisition is performed in a relaxed state where the child's palm is naturally extended and the five fingers are together; the second acquisition is performed in a stretched state where the child's palm is forcefully extended and the five fingers are maximally spread out; and the third acquisition is performed when the child's hand grips a grip strength meter and continuously exerts force until a preset grip strength threshold is reached; a preprocessing module, which preprocesses the spectral data acquired in the three acquisitions, including noise reduction, baseline correction, and normalization; and a feature extraction module. The system comprises four modules: a module for extracting trace element features related to children's health from preprocessed spectral data, including iron, zinc, calcium, magnesium, copper, and selenium; a feature vector construction module for constructing feature vectors based on data from three acquisitions, including static features corresponding to each acquisition and dynamic response features between different acquisition states; a model training module for training the feature vectors using a reinforcement learning-based artificial intelligence model, using historical datasets labeled with diseases; a data classification module for classifying new pediatric spectral data using the trained artificial intelligence model, outputting disease categories and probability distributions; and a risk assessment module for conducting risk assessments based on the classification results and generating risk assessment reports. This invention innovatively employs multi-state spectral acquisition technology combined with a reinforcement learning artificial intelligence model to achieve accurate and non-invasive classification and risk assessment of childhood diseases. By acquiring spectral data under three different physiological states, more comprehensive biometric information can be obtained. Dynamic response feature analysis enhances the sensitivity and specificity of disease identification. The reinforcement learning-based classification model can adaptively optimize classification strategies, improving diagnostic accuracy. The entire system automates the entire process from data acquisition to risk assessment, significantly improving the efficiency and reliability of childhood disease screening, providing effective technical support for early diagnosis and intervention, and possessing significant clinical application value and social benefits.

[0036] This invention captures dynamic response information of tissues under static, micro-change, and high-load conditions by acquiring spectral data under three different physiological states (relaxation, stretching, and loading), greatly enriching the feature dimensions and providing more sufficient evidence for disease identification. Specifically, this application achieves higher classification accuracy, and the extracted dynamic response features are more sensitive to physiological functional abnormalities under pathological conditions. Combined with a reinforcement learning-based AI model, it can autonomously learn the optimal classification strategy from high-dimensional complex features, significantly improving the accuracy, recall, and generalization ability of disease classification. The entire detection process is completely non-invasive, avoiding the pain and risks associated with radiation and blood sampling, making it particularly suitable for early screening and regular monitoring of children. The method has a high degree of standardization, facilitating its application in medical institutions at different levels. The final output includes not only simple classification results but also probability distributions and detailed risk assessment reports, providing clinicians with quantitative decision support and contributing to early disease warning and personalized intervention.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for classifying childhood diseases based on artificial intelligence, characterized in that, Includes the following steps: Step S1: Collect spectral data of the child's palm skin in three non-invasive ways under different physiological conditions using a spectrometer. The first collection is performed when the child's palm is naturally straight and the five fingers are together in a relaxed state. The second collection is performed when the child's palm is forcefully straight and the five fingers are spread out to the maximum extent. The third collection is performed when the child's hand is gripping the grip strength meter and continuously exerting force until the preset grip strength threshold is reached. Step S2: Preprocess the spectral data acquired in the three acquisitions, including noise reduction, baseline correction and normalization. Step S3: Extract trace element features related to children's health from the preprocessed spectral data, wherein the trace elements include iron, zinc, calcium, magnesium, copper and selenium; Step S4: Construct a feature vector based on the data from the three acquisitions, including the static features corresponding to each acquisition and the dynamic response features between different acquisition states; Step S5: Train the feature vector using an artificial intelligence model based on reinforcement learning, using a historical dataset with disease labels. Step S6: Use the trained artificial intelligence model to classify the new pediatric spectral data and output the disease categories and probability distributions; Step S7: Conduct a risk assessment based on the classification results and generate a risk assessment report.

2. The method for classifying childhood diseases based on artificial intelligence according to claim 1, characterized in that, The spectral data collected in the three sessions covered the visible to near-infrared bands, with a wavelength range of 400 nm to 2500 nm. During the collection, standardized protocols were used to control environmental conditions, including temperature, humidity, and light intensity. The preset grip strength threshold of the grip strength meter was dynamically adjusted according to the child's age, gender, and weight percentile to ensure the standardization and comparability of the load-bearing data.

3. The method for classifying childhood diseases based on artificial intelligence according to claim 2, characterized in that, The denoising process in the preprocessing step employs a moving average filtering method or a wavelet transform denoising algorithm; the baseline correction uses a polynomial fitting algorithm with a goodness of fit of not less than 0.95; and the normalization process uses a min-max normalization method or a z-score standardization method to ensure uniform data scale.

4. The method for classifying childhood diseases based on artificial intelligence according to claim 3, characterized in that, The extraction of trace element characteristics is achieved by analyzing the absorption peak, reflectance curve, and spectral integral area at a specific wavelength; the dynamic response characteristics include the difference, ratio, differential spectrum, or rate of change of integral area between any two data acquisitions.

5. The method for classifying childhood diseases based on artificial intelligence according to claim 4, characterized in that, The reinforcement learning-based artificial intelligence model adopts a deep reinforcement learning network architecture, including an actor-critic architecture or a deep Q-network; During training, the feature vector is used as the state input, the disease classification result is used as the action space, and the reward function guides the model optimization.

6. The method for classifying childhood diseases based on artificial intelligence according to claim 5, characterized in that, The training process in step S5 also includes an experience replay mechanism and a target network update strategy to improve training stability and convergence speed; the reward function is set according to the classification accuracy, with positive rewards for correct classification and negative rewards for incorrect classification.

7. The method for classifying childhood diseases based on artificial intelligence according to claim 6, characterized in that, The disease categories output in step S6 include health, tumor, intellectual disability, and ADHD; tumor is further subdivided into malignant tumor and benign tumor, and intellectual disability and ADHD are classified into mild, moderate and severe according to their severity.

8. The method for classifying childhood diseases based on artificial intelligence according to claim 7, characterized in that, The risk assessment step assesses the probability of a child contracting the disease by calculating a risk score, which is based on categorical probability, disease weighting factors, and clinical adjustment items. The risk assessment report includes the risk level, recommended measures, and confidence level indicators.

9. A method for classifying childhood diseases based on artificial intelligence according to claim 8, characterized in that, It also includes model validation and optimization steps: evaluating model performance using independent test datasets, with evaluation metrics including accuracy, recall, F1 score, and AUC; and optimizing model parameters through cross-validation and hyperparameter tuning.

10. A child disease classification system based on artificial intelligence, characterized in that, The method for classifying childhood diseases based on artificial intelligence as described in any one of claims 1-9 includes: The data acquisition module uses a spectrometer to collect spectral data of children's palm skin in three non-invasive ways under different physiological conditions. The first acquisition is carried out in a relaxed state with the child's palm naturally straight and the five fingers together. The second acquisition is carried out in a stretched state with the child's palm forcefully straight and the five fingers spread out to the maximum extent. The third acquisition is carried out when the child's hand holds the grip strength meter and continues to exert force until the preset grip strength threshold is reached. The preprocessing module performs preprocessing on the spectral data acquired in the three acquisitions, including noise reduction, baseline correction, and normalization. The feature extraction module extracts trace element features related to children's health from the preprocessed spectral data. The trace elements include iron, zinc, calcium, magnesium, copper, and selenium. The feature vector construction module constructs feature vectors based on data collected from three sampling sessions, including static features corresponding to each sampling session and dynamic response features between different sampling states. The model training module uses a reinforcement learning-based artificial intelligence model to train the feature vectors, and the training uses a historical dataset with disease labels. The data classification module uses a trained artificial intelligence model to classify new pediatric spectral data and output disease categories and probability distributions. The risk assessment module performs risk assessments based on the classification results and generates risk assessment reports.

Images