Information processing method, information processing device, program, and recording medium
The technology integrates clustering and statistical modeling to identify causal skin relationships, offering personalized skincare solutions by analyzing multiple skin factors.
Patent Information
- Application Number
- JP2025077777
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-05-08
- Publication Date
- 2026-11-19
AI Technical Summary
Conventional methods struggle to clearly identify causal relationships between multiple pieces of skin information, failing to integrate and evaluate skin conditions effectively.
An information processing technology that utilizes clustering and statistical modeling to estimate causal relationships between skin function and phenotype information, incorporating confounding factors like sex, age, and lifestyle, to generate personalized skincare recommendations.
Enables the estimation of causal relationships between skin conditions, providing personalized skincare interventions based on statistically significant associations.
Smart Images

Figure 2026181940000001_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present invention relates to an information processing technology for analyzing biological information, particularly information related to the skin. BACKGROUND ART
[0002] Technologies exist for evaluating an individual's skin condition and proposing appropriate care methods and the like. In such technologies, analysis of the relationship between a plurality of measured pieces of information related to the skin is performed. For example, there is also known a technology that acquires specific color information (for example, redness information) obtained from the skin of a subject, and uses that information to estimate internal body factors that may be related to skin problems (see, for example, Patent Document 1). PRIOR ART DOCUMENTS PATENT DOCUMENTS
[0003] PATENT DOCUMENT 1 Japanese Unexamined Patent Application Publication No. 2018-102730 SUMMARY OF INVENTION PROBLEMS TO BE SOLVED BY THE INVENTION
[0004] However, in conventional methods for analyzing information related to the skin, although attempts have been made to find associations between different pieces of information, it has not necessarily been easy to clearly identify even causal relationships such as whether the association is a mere correlation or whether one is the cause of the other. In addition, while the skin condition is considered to involve not only a single piece of information but also a plurality of functional pieces of information in a complex manner, approaches that handle these plurality of functional information in an integrated manner to evaluate the skin condition and further estimate the causal relationship with phenotypes have not been sufficient.
[0005] The present invention has been made in view of the above circumstances, and a main object thereof is to provide an information processing technology capable of estimating causal relationships between information related to the skin condition using a plurality of pieces of information related to the skin. MEANS FOR SOLVING THE PROBLEMS
[0006] The present invention Computers An input step in which at least three different types of skin function information and skin phenotype information are entered, A profile creation step involves performing clustering using the aforementioned skin function information to create a skin function profile, For each skin function profile, an incidence index estimation step is performed to estimate an incidence index indicating the likelihood of the occurrence of the skin phenotypic information using a statistical model. The present invention provides an information processing method that performs an estimation step of estimating the causal relationship between the skin function profile and the skin phenotype information using the statistical model, based on the occurrence index estimated in the occurrence index estimation step and the skin function profile. The estimation step may involve using a regression model to estimate that the skin function profile causally contributes to the skin phenotypic information if the confidence interval of the regression coefficient corresponding to the variable representing the skin function profile does not contain zero at a predetermined confidence level. The aforementioned confidence interval may be calculated based on a 95% confidence level. The clustering may be performed using one of the following methods based on the input skin function information: k-means, hierarchical clustering, and a Gaussian mixture model. The regression model may include one of the following confounding factors as a confounding factor when estimating the relationship between the skin function profile and the skin phenotypic information: sex, age, environmental factors, and lifestyle. The computer may further perform a recommendation information generation step to generate recommendation information for improving the skin phenotype information based on the causal relationships estimated in the estimation step. The aforementioned recommended information may include information on at least one selected from the group consisting of skincare habits, lifestyle habits, skincare products, quasi-drugs, supplements, functional foods, and beauty devices. Furthermore, the present invention is An input unit that inputs at least three different types of skin function information and skin phenotype information, A profile creation unit that performs clustering using the aforementioned skin function information and creates a skin function profile, For each skin function profile, an occurrence index estimation unit estimates an occurrence index indicating the likelihood of the occurrence of the skin phenotypic information using a statistical model, The present invention provides an information processing device comprising: an estimation unit that estimates the causal relationship between the skin function profile and the skin phenotype information using the statistical model, based on the occurrence index estimated by the occurrence index estimation unit and the skin function profile. Furthermore, the present invention is On the computer, An input step in which at least three different types of skin function information and skin phenotype information are entered, A profile creation step involves performing clustering using the aforementioned skin function information to create a skin function profile, For each skin function profile, an incidence index estimation step is performed to estimate an incidence index indicating the likelihood of the occurrence of the skin phenotypic information using a statistical model. The present invention provides a program for performing an estimation step, which involves estimating the causal relationship between the skin function profile and the skin phenotypic information using the statistical model, based on the incidence index estimated in the incidence index estimation step and the skin function profile. Furthermore, the present invention is The present invention provides a computer-readable recording medium on which the aforementioned program is stored. [Effects of the Invention]
[0007] According to the present invention, it is possible to estimate the causal relationships between pieces of information related to the condition of the skin by using multiple pieces of information about the skin. [Brief explanation of the drawing]
[0008] [Figure 1]FIG. 1 is a block diagram illustrating an example of a hardware configuration of an information processing apparatus according to an embodiment of the present invention. [Figure 2] FIG. 2 is a block diagram illustrating an example of a functional configuration of the information processing apparatus according to an embodiment of the present invention. [Figure 3] FIG. 3 is a flowchart illustrating an example of a processing procedure of an information processing method according to an embodiment of the present invention. [Description of Embodiments]
[0009] Hereinafter, preferred embodiments for carrying out the present invention will be described in detail with reference to the drawings. The embodiments described below are merely illustrative of representative embodiments of the present invention, and the scope of the present invention is not limited thereby. Further, the present invention can be combined with any of the following examples and modifications thereof.
[0010] In the following description of the embodiments, the configuration may be described using terms accompanied by "substantially", such as substantially parallel and substantially orthogonal. For example, "substantially parallel" not only means perfectly parallel, but also means substantially parallel, that is, including a state shifted by, for example, about several percent from a perfectly parallel state. The same applies to other terms accompanied by "substantially".
[0011] Each drawing is a schematic diagram and is not necessarily strictly illustrated. The scales in the drawings are emphasized to make the technical features easy to understand. Therefore, it should be noted that the scales in the drawings are not necessarily the same as the scales of actual devices.
[0012] Unless otherwise specified, in the drawings, "upper" means the upward direction or upper side in the drawings, "lower" means the downward direction or lower side in the drawings, "left" means the left direction or left side in the drawings, and "right" means the right direction or right side in the drawings. In the drawings, the same or equivalent elements or members are denoted by the same reference numerals, and redundant descriptions are omitted.
[0013] Note that the effects described in the present specification are merely illustrative and not restrictive, and other effects may also be provided.
[0014] The description in this specification will be given in the following order. 1. Configuration of Information Processing Apparatus 1-1. Overall Configuration 1-2. Functional Configuration 1-2-1. Input Unit 201 1-2-2. Profile Creation Unit 202 1-2-3. Occurrence Indicator Estimation Unit 203 1-2-4. Estimation Unit 204 1-2-5. Recommendation Information Generation Unit 205 2. Information Processing Method 2-1. Overall Processing Flow 2-2. Input Step: S101 2-3. Profile Creation Step: S102 2-4. Occurrence Indicator Estimation Step: S103 2-5. Estimation Step: S104 2-6. Recommendation Information Generation Step: S105 3. Summary of Functions and Effects 4. Modified Examples 5. Program, Recording Medium
[0015] <1. Configuration of Information Processing Apparatus> First, the configuration of the information processing apparatus according to an embodiment of the present invention will be described. FIG. 1 is a block diagram showing an example of the hardware configuration of the information processing apparatus 100 according to the present embodiment.
[0016] (1-1. Overall Configuration) The information processing device 100 can be configured as, for example, a personal computer, a server computer, a tablet terminal, a smartphone, or a dedicated analysis device. As shown in Figure 1, the information processing device 100 includes a control unit (CPU: Central Processing Unit) 101, main memory (RAM: Random Access Memory) 102, auxiliary storage (e.g., HDD: Hard Disk Drive, SSD: Solid State Drive) 103, an input interface 104, an output interface 105, and a communication interface 106, which are interconnected via a bus 107.
[0017] The control unit 101 controls the operation of the entire information processing device 100 by reading programs (including information processing programs described later) stored in the auxiliary storage device 103 into the main storage device 102 and executing them, and also implements various functions described later (input unit, profile creation unit, occurrence indicator estimation unit, estimation unit, recommendation information generation unit, etc.).
[0018] The main memory 102 functions as a work memory that temporarily stores programs executed by the control unit 101 and data being processed.
[0019] The auxiliary storage device 103 stores the OS (Operating System), various application programs, and data to be processed (skin function information, skin phenotype information, generated profiles, estimation results, recommendation information database, etc.). The information processing program according to the present invention can also be stored in this auxiliary storage device 103.
[0020] The input interface 104 accepts data input from a keyboard, mouse, touch panel, scanner, camera, or external measuring device (such as a skin measuring device). Skin function information and skin phenotype information can be input via this input interface 104. The input data format is not particularly limited and can be various formats such as numerical data, text data, or image data. The input data may be preprocessed as needed (e.g., imputation of missing values, noise reduction, format conversion, etc.).
[0021] The output interface 105 is connected to a display, printer, or other external display device, and outputs the processing results (estimated causal relationships, generated recommendation information, etc.).
[0022] The communication interface 106 transmits and receives data to and from external database servers and other computers via a network (e.g., LAN: Local Area Network, Internet, etc.). For example, it is used to retrieve skin information from an external database or to send processing results to an external system.
[0023] (1-2. Functional Configuration) Next, the functional configuration of the information processing device 100 will be described. Figure 2 is a block diagram showing an example of the functional configuration of the information processing device 100 according to this embodiment. The information processing device 100 functions as an input unit 201, a profile creation unit 202, an occurrence indicator estimation unit 203, an estimation unit 204, and a recommendation information generation unit 205, when the control unit 101 executes a predetermined information processing program. These functional units can be implemented by hardware, software, or a combination thereof.
[0024] (1-2-1. Input section 201) The input unit 201 has the function of inputting data necessary for information processing. Specifically, it receives at least three different types of skin function information and skin phenotype information via the input interface 104 and the communication interface 106, and stores them in the main memory 102 and the auxiliary memory 103.
[0025] The "skin function information" that is input refers to information about the physiological function of the skin, and is quantitative or qualitative data obtained by various measuring instruments or biochemical analyses. The types are not particularly limited, but examples include stratum corneum water content, transepidermal water loss (TEWL), dermal water content, sebum amount, sebum composition ratio, skin surface pH, skin elasticity, skin color (melanin index, hemoglobin index, etc.), skin temperature, state of stratum corneum cells (area, overlap, exfoliation, etc.), skin thickness, collagen amount, elastin amount, expression level of specific proteins (e.g., desmoglein amount), information on the skin's resident microbiota (total amount, amount of each bacterial species, bacterial species composition, diversity, etc.), and genetic information (SNP: Single Nucleotide Polymorphism, etc.).
[0026] In this invention, it is preferable from the viewpoint of creating a stable profile to input at least three different types of information from among this information. Preferably, five or more types of skin function information are used, and more preferably, ten or more types.
[0027] Because this information is susceptible to individual differences and measurement conditions, standardizing the measurement environment (temperature, humidity, etc.), measurement site, and measurement procedure is crucial for improving the reliability of the analysis results.
[0028] The "skin phenotypic information" entered refers to information about the appearance and morphology of the skin, as well as subjective symptoms, and is quantitative or qualitative data obtained through visual inspection, palpation, expert evaluation, image analysis, questionnaires, etc. Examples include wrinkles (crow's feet, nasolabial folds, etc.), age spots (senile lentigines, melasma, etc.), skin texture, pore condition (enlarged, blackheads, etc.), acne, redness, dullness, skin brightness, transparency, skin health, sensitivity (irritation, itchiness, etc.), and the presence and degree of specific skin diseases (seborrheic dermatitis, rosacea, etc.).
[0029] (1-2-2. Profile Creation Section 202) The profile creation unit 202 has the function of creating a skin function profile by performing clustering using multiple skin function information input by the input unit 201.
[0030] Clustering is an unsupervised learning method that divides data points into groups (clusters) based on similarity. The profile creation unit 202 treats the input individual's multiple skin function information as points in a multidimensional space, and based on the distribution of these points, it classifies individuals with similar skin function patterns into the same cluster.
[0031] This clustering can be performed, for example, using the k-means method. The k-means method minimizes the sum of squares within a cluster by repeatedly assigning data to the nearest cluster center (centroid) from each data point, based on a predetermined number of clusters k, and recalculating the cluster centers.
[0032] Methods for determining the number of clusters k include, for example, the elbow method (a method for finding the point where the rate of change of the sum of squares within a cluster becomes gradual), silhouette analysis, and gap statistics. In this embodiment, the optimal number of clusters is determined empirically or statistically depending on the type and distribution of the input skin function information.
[0033] Clustering methods are not limited to the k-means method; other methods may also be used. For example, hierarchical clustering (agglomerative and partitioning) and Gaussian mixture models (GMMs) can also be effective.
[0034] Hierarchical clustering is a method that calculates the similarity (or dissimilarity) between data points and then gradually merges the closest data points or clusters based on that similarity. It can be done in an agglomerative (bottom-up) or, conversely, a partitioning (top-down) method that divides the data from the whole. Advantages of this method include the fact that the number of clusters does not need to be specified in advance, and the hierarchical structure of the data can be visually confirmed using a dendrogram (tree diagram). By cutting the dendrogram at appropriate heights, the desired number of clusters can be obtained.
[0035] A Gaussian mixture model (GMM) is a probabilistic clustering method that assumes observed data is generated from a mixture of multivariate Gaussian distributions (normal distributions) with multiple different means and covariance matrices. Using algorithms such as EM (Expectation-Maximization), it estimates the posterior probability that each data point belongs to a particular Gaussian distribution (cluster). This allows for not only assigning a single data point to the cluster with the highest probability (hard clustering), but also soft clustering, which shows the degree to which a data point belongs to multiple clusters. Furthermore, while the k-means method mainly assumes spherical clusters, GMM has the advantage of being able to flexibly model elliptic clusters by considering the covariance matrix.
[0036] In addition to these methods, DBSCAN (Density-Based Clustering) is another option. Since the definition of similarity and the shape of clusters differ depending on the method used, it is desirable to select an appropriate method based on the characteristics of the data.
[0037] A "skin function profile" refers to each cluster obtained through the clustering described above, i.e., a group of individuals with similar skin function patterns. Each profile (cluster) is characterized by a specific combination of skin functions. For example, it could be defined as "Profile 1: A group with high moisture content but low barrier function," or "Profile 2: A group with high sebum content but standard moisture content."
[0038] The profile creation unit 202 generates information indicating which skin function profile each individual belongs to. The number of profiles is determined by the clustering results, but at least two or more are created.
[0039] (1-2-3. Occurrence Indicator Estimation Unit 203) The occurrence index estimation unit 203 has the function of estimating occurrence indices, which indicate the likelihood of occurrence of the input skin phenotype information, for each skin function profile created by the profile creation unit 202, using a statistical model.
[0040] An "occurrence index" is an indicator that shows how likely or at what level a particular skin phenotype (e.g., a particular skin problem) is to occur in an individual belonging to a specific skin function profile. Different values are used depending on the nature of the skin phenotype information being studied (quantitative / qualitative, continuous / categorical variable). For example, it could be the frequency (incidence rate) of a particular skin problem (e.g., acne), the mean or median of a particular phenotype (e.g., wrinkle level), or the risk of a particular phenotype occurring relative to a baseline profile (e.g., odds ratio, risk ratio).
[0041] A "statistical model" is a general term for methods that mathematically model the relationships between data. The incidence index estimation unit 203 constructs a statistical model to analyze the relationship between the skin function profile and skin phenotypic information, and uses it to estimate the incidence index.
[0042] A "regression model" can be used as this statistical model. A regression model is a model that predicts or explains a dependent variable (output) using explanatory variables (inputs). In this step, the dependent variable is specific skin phenotype information, and skin function profiles are included as explanatory variables. In this case, each profile is treated as a categorical variable, and by converting them into dummy variables for a baseline profile, for example, and inputting them into the model, it becomes possible to individually estimate and evaluate the influence of each skin function profile on skin phenotype information.
[0043] When analyzing the relationship between skin function profiles and skin phenotypic information using statistical models (especially regression models), it is preferable to include "confounding factors" that may influence both in the model. Confounding factors are variables that relate to both the explanatory and dependent variables, creating apparent relationships or masking true relationships.
[0044] While a wide range of confounding factors influence human skin condition, sex and age, in particular, affect many skin functions and phenotypes. Therefore, including these as confounding factors in the model is expected to allow for a more accurate estimation of the relationship between the profile and phenotype.
[0045] Furthermore, skin condition is greatly influenced by "environmental factors" surrounding an individual (e.g., sunlight hours, climate, season, daily UV exposure levels, humidity in living spaces, etc.) and "lifestyle habits" (e.g., diet, sleep, exercise, stress management, smoking and drinking habits, etc.). These factors can also be related to both skin function profiles and skin phenotypic information, and therefore can be important confounding factors.
[0046] In this invention, in order to appropriately adjust for the influence of the confounding factors described above, it is effective to incorporate these factors into the model based on the data available for analysis. Specifically, at least one of either sex or age, preferably both, should be included in the model. Furthermore, if available, adding information on environmental factors and lifestyle can contribute to a more precise estimation of causal relationships. In other words, one of the group consisting of sex, age, environmental factors, and lifestyle should be included in the regression model as a confounding factor. Other candidate confounding factors include, for example, race, genetic factors, disease, and psychological state, and if this information is available, including it in the model is also an effective modification.
[0047] The type of regression model used is selected according to the nature of the skin phenotype information, which is the dependent variable. For example, if the phenotype is binary (e.g., presence or absence of acne), a logistic regression model is used; if it is continuous (e.g., wrinkle score), a linear regression model is used; and if it is ordinal categorical (e.g., skin brightness rating), an ordinal logistic regression model is used. In addition, if there are many explanatory variables or if multicollinearity is a concern, it is also effective to use regularized regression models such as Lasso (Least Absolute Shrinkage and Selection Operator) regression or Ridge regression.
[0048] Statistical models are not limited to regression models; other machine learning models (e.g., decision trees, random forests, gradient boosting, support vector machines, neural networks, etc.) can also be used. These models are often better suited to capturing nonlinear relationships and complex interactions between variables.
[0049] (1-2-4. Estimation section 204) The estimation unit 204 has the function of estimating the causal relationship between the skin function profile and skin phenotypic information using a statistical model (the one used in the incidence index estimation step, or a related model) based on the incidence index estimated by the incidence index estimation unit 203 and the skin function profile.
[0050] In this step, "estimating causality" means statistically evaluating, based on observational data, the possibility that a specific skin function profile (having a specific functional pattern) influences (causes) the probability or severity of a specific skin phenotype (such as skin problems). This goes beyond mere correlation (tendencies to fluctuate simultaneously) and suggests a directional relationship where one causes the other. However, since estimating causality from observational study data has limitations due to the presence of unmeasured confounding factors, the estimation results should be interpreted not as proof of definitive causality, but as a statistically supported possibility of causal association.
[0051] The estimation unit 204 estimates causal relationships using the results of the statistical model, particularly the regression model, constructed and used by the occurrence index estimation unit 203. In this regression model, it evaluates the "regression coefficients" and their "confidence intervals" corresponding to variables that represent a specific skin function profile (e.g., dummy variables for the reference profile). The regression coefficients indicate the magnitude (including positive and negative directions) of the influence that profile has on the skin phenotype, which is the dependent variable. The confidence interval indicates the range of statistical uncertainty of the estimated regression coefficients.
[0052] If this confidence interval does not contain 0 at a given confidence level, it is estimated that the skin function profile "causally contributes" to the skin phenotypic information (i.e., has a statistically significant effect). Here, it is preferable to use "95%" as the "given confidence level." A 95% confidence interval that does not contain 0 means that the null hypothesis that "the influence of the profile is zero" is rejected at a significance level of 5%, suggesting a statistically significant association (possibility of causality). If the confidence interval contains 0, it cannot be determined that there is a causal contribution, at least statistically. The confidence level is not limited to 95%; other levels such as 90% or 99% can be used depending on the purpose.
[0053] Furthermore, it is common practice to evaluate the p-value of the regression coefficient (the probability of obtaining the observed result or an even more extreme result under the null hypothesis) and determine that there is a significant association if the p-value is less than a predetermined significance level (e.g., 0.05). Based on these statistical indicators, the estimation unit 204 estimates whether there is a causal relationship between each skin function profile and each skin phenotypic information, and whether that relationship increases or decreases the risk. Note that the p-value and confidence interval are complementary evaluation indicators, and the validity of the interpretation is enhanced when considered together with the width and inclusion relationship of the confidence interval.
[0054] While the goodness of fit of a statistical model (e.g., coefficient of determination, AIC: Akaike Information Criterion, BIC: Bayesian Information Criterion) may be considered when evaluating a statistical model, the primary objective of this step is to estimate causal relationships, so more emphasis is placed on the significance and confidence intervals of regression coefficients than on predictive accuracy. However, if the model fits extremely poorly, the reliability of the estimated causal relationships may also decrease, so it is desirable to confirm the validity of the model to a certain extent.
[0055] (1-2-5. Recommended Information Generation Unit 205) The recommendation information generation unit 205 has the function of generating recommendation information to improve specific skin phenotype information based on the causal relationships estimated by the estimation unit 204.
[0056] Specifically, the process begins by identifying the skin phenotype to be improved (for example, skin problems that the user is concerned about, or phenotypes that are estimated to be at high risk). Next, it identifies the skin function profile that is estimated to causally contribute to (increase the risk of) that phenotype. Finally, considering the characteristics of that profile (for example, low or high levels of specific functions), it generates recommendations on intervention methods that are considered effective for individuals with that profile to improve the target phenotype.
[0057] The types of recommended information include, for example, "skincare habits" (e.g., face washing methods, moisturizing methods, UV protection, products used, etc.), "lifestyle habits" (e.g., diet, sleep, exercise, stress management, smoking / drinking habits, etc.), "skincare products" (e.g., lotions, serums, creams containing specific ingredients, etc.), "quasi-drugs" (e.g., topical products containing active ingredients with whitening or anti-inflammatory effects, etc.), "functional foods" (e.g., supplements, beverages expected to improve skin function, etc.), and "beauty devices" (e.g., ultrasonic devices, LED irradiators, iontophoresis devices, etc.). These are selected in a way that addresses the cause based on the estimated causal relationship (e.g., "This profile is prone to redness due to a weakened barrier function caused by dryness"). For example, if a weakened barrier function is the cause, then moisturizers containing ceramides or skincare habits that emphasize moisturizing would be recommended.
[0058] The recommendation information generation unit 205 may refer to internal or external databases (e.g., a skincare product database, a food ingredient database, a beauty information database, etc.) to generate this recommendation information. The generated recommendation information is presented to the user in an easily understandable format (e.g., text, graphics, reports, etc.) via the output interface 105.
[0059] <2. Information Processing Methods> Next, the information processing method executed by the information processing device 100 with the above configuration will be explained with reference to the flowchart example in Figure 3. Figure 3 is a flowchart of an example of the processing procedure of the information processing method according to this embodiment.
[0060] (2-1. Overall processing flow) As shown in Figure 3, the information processing method of this embodiment mainly includes an input step (S101), a profile creation step (S102), an occurrence indicator estimation step (S103), an estimation step (S104), and optionally a recommendation information generation step (S105).
[0061] (2-2. Input Step: S101) First, the input unit 201 receives information about the individual's skin to be processed (S101). As mentioned above, the information to be input consists of at least three different types of "skin function information" and "skin phenotype information." This information can be input in various ways, such as by direct input by the user, acquisition from an external measuring device, or reading from a database.
[0062] For example, in dermatology clinics or beauty salons, specialized equipment such as a corneometer (stratum corneum moisture content), Tevameter (TEWL), Cutometer (viscoelasticity), sebum meter, and melanin / hemoglobin meter are used to measure skin function information, and at the same time, a camera (e.g., VISIA®) is used to capture facial images and acquire skin phenotypic information such as wrinkles and blemishes, and this data is input into the information processing device 100. Alternatively, it is also conceivable to continuously collect and input skin information using wearable sensors or the like.
[0063] The input data is preprocessed as needed (unit unification, outlier removal, normalization, etc.) to prepare it for analysis. This preprocessing is important because it affects the accuracy of subsequent steps.
[0064] This step prepares the basic data necessary for analysis, and has the effect of building an information infrastructure that allows for a multifaceted understanding of an individual's detailed skin condition.
[0065] (2-3. Profile creation step: S102) Next, the profile creation unit 202 performs clustering using the multiple skin function information input in the input step (S101) to create a skin function profile (S102).
[0066] Specifically, the individual skin function information vectors are plotted in a multidimensional space, and clustering algorithms such as k-means are applied to group individuals with similar function patterns. For example, if data for N individuals is available, each individual is assigned to one of k profiles (where k is predetermined, e.g., 5).
[0067] In this step, it is preferable to use at least one of the following: k-means, hierarchical clustering, and Gaussian mixture models. However, other clustering methods (such as DBSCAN) may be used depending on the characteristics of the data. Using these alternative methods may reveal clusters of different shapes. When using k-means, the number of clusters k can be determined using model selection criteria such as the elbow method, silhouette score, or gap statistic.
[0068] Furthermore, it is important that at least three types of skin function information are input in order to perform stable clustering and form a meaningful profile. By changing the types and number of functional information used, it may be possible to obtain profiles from different perspectives. For example, it is possible to create a profile using only moisture-related information, only sebum-related information, or a combination of these, depending on the purpose.
[0069] This step allows for the reduction and organization of complex, high-dimensional skin function information into a smaller, more easily interpretable profile (skin type). This has the effect of providing a comprehensive understanding of an individual's skin condition and facilitating subsequent causal relationship estimation. It also has the potential to discover previously unknown skin types. In addition, the resulting profiles can be used as input variables for machine learning and personalized recommendation systems.
[0070] (2-4. Estimation step for occurrence indicators: S103) Next, the occurrence index estimation unit 203 estimates occurrence indices (S103) that indicate the likelihood of occurrence of the input skin phenotypic information for each skin function profile obtained in the profile creation step (S102), using a statistical model.
[0071] Specifically, first, one piece of skin phenotypic information (for example, "presence or absence of blemishes" or "wrinkle score") is selected. Next, a statistical model (for example, a regression model) is constructed with that phenotypic information as the dependent variable and the skin function profile (information indicating which profile each individual belongs to, usually represented as a dummy variable) as one of the independent variables.
[0072] In this process, including confounding factors such as gender and age as explanatory variables in the model is crucial for improving the accuracy and validity of the estimation. This allows for the evaluation of profile-specific influences separately from those of age and gender. In addition to gender and age, other useful variations of confounding factors to include in the model include other available information (e.g., skincare habits, lifestyle habits).
[0073] After constructing the model, the model is used to calculate the predicted values of the dependent variable in each skin function profile, as well as the influence of a particular profile on the dependent variable (e.g., odds ratio or difference in mean values), and this is called the "occurrence index." This index is used to quantitatively understand the risk and severity of the occurrence of a particular phenotype in each profile. For example, if a logistic regression model is used, an odds ratio of 2.0 for a certain profile indicates that individuals belonging to that profile are twice as likely to develop the target phenotype (e.g., freckles) compared to the baseline profile.
[0074] By repeating this step for multiple skin phenotypic data points of interest, a comprehensive picture emerges of which skin expressions (such as problems) each profile is likely to be associated with. This result can be used to create risk maps and preventive recommendations for each profile.
[0075] This step allows for the estimation of quantitative relationships between each skin function profile and individual skin phenotypes. This has the effect of objectively evaluating what skin risks individuals with a particular profile are prone to.
[0076] (2-5. Estimation Step: S104) Next, the estimation unit 204 estimates the causal relationship between the skin function profile and skin phenotypic information using a statistical model, based on the results of the occurrence index estimation step (S103) and the skin function profile (S104).
[0077] In this step, the results of the statistical model (particularly the regression model) used in the occurrence indicator estimation step (S103) are further interpreted, focusing on the model's coefficient estimates and their uncertainties, and the likelihood of a causal relationship existing is evaluated based on statistical significance.
[0078] Specifically, we focus on the regression coefficients of the variables representing each skin function profile in the regression model (such as dummy variables for the baseline profile) and calculate their 95% confidence intervals. Then, we check whether these 95% confidence intervals cross zero (i.e., whether both the lower and upper limits are positive or both are negative). If the confidence interval does not contain zero, it is estimated that the profile has a statistically significant effect on the target skin phenotype, i.e., it is highly likely to "causally contribute". For example, if the confidence interval is [0.5, 1.5], it can be judged that there is a significant positive effect (increased risk), and if it is [-1.0, -0.2], it can be judged that there is a significant negative effect (decreased risk). If the confidence interval contains zero, such as [-0.3, 0.8], it is judged that there is no statistically significant effect. The confidence level is generally 95%, but it is possible to use other levels (90%, 99%, etc.) as needed. Alternatively, significance can be judged using p-values, with a criterion such as p<0.05. In this embodiment, a regression model was used, but it is possible to modify the model to use other causal estimation methods depending on the data structure and purpose.
[0079] This step allows for the estimation of a causal relationship (the possibility that the profile is the cause of the phenotype) based on statistical evidence, rather than mere correlation. This has the significant effect of identifying which skin function pattern (profile) is most likely to be the root cause of skin problems, and providing scientific evidence for considering more fundamental countermeasures. This step can serve as the basis for scientific evidence that contributes to the formulation of preventive and intervention measures. Importantly, it enables the estimation of the direction of cause and effect, which was difficult with conventional correlation analysis.
[0080] (2-6. Recommended information generation step: S105) Optionally, the recommendation information generation unit 205 generates recommendation information to improve the target skin phenotype information based on the causal relationship estimated in the estimation step (S104) (S105).
[0081] First, based on the estimated causal relationships, identify skin function profiles that are judged to increase the risk (or have room for improvement) of the skin phenotype to be improved. Next, considering the characteristics of that profile (which functions are high or low, etc.) and the estimated causal relationship mechanism, select approaches (skincare habits, lifestyle habits, products, foods, etc.) that may lead to improvement and generate specific recommendations.
[0082] For example, if a causal relationship is estimated that "Profile A (dryness / impaired barrier type) significantly increases the risk of wrinkles," then recommendations such as "use a highly moisturizing serum containing ceramides," "use a humidifier," and "drink plenty of water" may be generated. These recommendations are personalized according to the individual's profile and target phenotype.
[0083] The types of recommended information are diverse and may include skincare methods (cleansing, moisturizing, sun protection, etc.), lifestyle habits (diet, sleep, exercise, stress, etc.), specific products (cosmetics, supplements, etc.), or recommendations to consult with professionals (dermatologists, estheticians, etc.).
[0084] This step proposes specific, evidence-based improvement measures based on the underlying causes (estimated profiles) of the individual's skin condition. This allows users to efficiently perform effective care tailored to their needs, leading to improved quality of skincare management.
[0085] <3. Summary of Actions and Effects> According to the information processing method, information processing device, and program of this embodiment described above, it is possible to input multiple different types of skin function information, cluster them to create an individual skin function profile, and further estimate the causal relationship between this profile and skin phenotypic information using a statistical model.
[0086] This makes it possible to statistically clarify the cause-and-effect relationships between various pieces of information about skin condition, which was previously difficult. For example, it can identify which combination pattern (profile) of skin function is most likely to be the cause of a particular skin problem.
[0087] As a result, it becomes possible to gain a deeper understanding of skin condition and select and develop effective care methods based on the underlying causes. Furthermore, by generating personalized recommendations based on estimated causal relationships, it is possible to provide skincare management solutions optimized for each user's individual skin condition. This enables a cyclical skincare support system that goes beyond simple skin condition monitoring, including the suggestion of personalized interventions and feedback on their evaluation.
[0088] Thus, this embodiment provides useful knowledge and tools in a wide range of fields, including dermatological research, cosmetic development, beauty counseling, and personal healthcare.
[0089] The effects described herein are merely illustrative and not limiting, and other effects may also occur. For example, the profiles and causal relationships obtained by this invention can be applied to the development of new skin evaluation indices and the design of clinical studies targeting populations with specific profiles. Furthermore, the method of this invention is not limited to skin but has applicability to health prediction and recommendations based on other biological information. It is sufficient to achieve any one of the multiple effects described (e.g., cause identification, appropriate suggestions, etc.).
[0090] <4. Variation> The present invention is not limited to the embodiments described above, and various modifications are possible without departing from the spirit of the invention.
[0091] For example, the types and number of skin function information inputs can be changed according to the purpose. You can focus on specific functions (e.g., moisture, sebum, barrier function) or integrate a wider variety of information (e.g., genetic information, microbiome information, lifestyle information).
[0092] Skin phenotypic information may include not only visual evaluations but also various aspects such as the user's subjective symptoms and satisfaction level.
[0093] Other clustering methods besides k-means, hierarchical clustering, and Gaussian mixture models (such as density-based clustering) may be applied, or multiple methods may be combined to validate the profile.
[0094] In addition to regression models, other statistical models such as structural equation modeling (SEM), Bayesian networks, or various machine learning models (random forests, gradient boosting, neural networks, etc.) can also be used. These methods can be effective in capturing more complex relationships and nonlinearities between variables.
[0095] In addition to sex, age, environmental factors, and lifestyle, incorporating confounding factors such as race, genetic factors, disease, and psychological state into the model can sometimes enable a more precise estimation of causal relationships.
[0096] In the recommendation information generation step, the types and methods of presenting recommendations can also be varied in many ways. For example, it is possible to use AI (artificial intelligence) to generate more personalized advice, or to provide information interactively in a chatbot format.
[0097] The system configuration of the information processing device may be provided not only as a standalone device, but also as a client-server system or a cloud-based service. For example, it is possible for a user to input skin information using a smartphone app, for analysis processing to be performed on a server in the cloud, and for the results to be returned to the app.
[0098] Furthermore, each step and each functional unit of the above embodiment can be combined arbitrarily. For example, a configuration in which the causal relationship estimation (S104) is performed without executing the recommendation information generation step (S105) is also within the scope of the present invention. Similarly, it is possible to employ an algorithm that integrates the occurrence indicator estimation step (S103) and the estimation step (S104).
[0099] It is also possible to combine components described in different embodiments (for example, a specific clustering method and a specific statistical model). For example, if the k-means method is used in Embodiment 1 and hierarchical clustering is used in Embodiment 2, it is conceivable to perform an analysis combining both. In particular, the combination of the clustering method used in the profiling step (e.g., k-means method, hierarchical clustering) and the statistical model used in the occurrence indicator estimation / estimation step (e.g., linear regression, logistic regression, machine learning model) can be selected in various ways depending on the characteristics of the data and the purpose of the analysis, and a particular combination (for example, the combination of k-means method and logistic regression) may be particularly effective under certain circumstances.
[0100] <5. Programs, recording media> Each step of the information processing method described in the above embodiment can be implemented as a program to be executed by a computer. This program can be provided by recording it on a computer-readable recording medium (e.g., CD-ROM, DVD, semiconductor memory, hard disk, etc.). The program can also be downloaded via a network. The computer loads this program into its main memory and executes it, thereby realizing the functions of the input unit 201, profile creation unit 202, occurrence indicator estimation unit 203, estimation unit 204, and recommendation information generation unit 205 described above, and performing the information processing shown in Figure 3. In other words, the present invention can also be understood as the program and the computer-readable recording medium on which the program is recorded.
[0101] Furthermore, the present invention can also take the following configuration. (1) Computers An input step in which at least three different types of skin function information and skin phenotype information are entered, A profile creation step involves performing clustering using the aforementioned skin function information to create a skin function profile, For each skin function profile, an incidence index estimation step is performed to estimate an incidence index indicating the likelihood of the occurrence of the skin phenotypic information using a statistical model. An information processing method that performs an estimation step of estimating the causal relationship between the skin function profile and the skin phenotype information using the statistical model, based on the occurrence index estimated in the occurrence index estimation step and the skin function profile. (2) The estimation step involves using a regression model to estimate that the skin function profile causally contributes to the skin phenotypic information if the confidence interval of the regression coefficient corresponding to the variable representing the skin function profile does not contain zero at a predetermined confidence level. The information processing method described in (1) above. (3) The aforementioned confidence interval was calculated based on a 95% confidence level. The information processing method described in (2) above. (4) The clustering is performed using one of the following methods based on the input skin function information: k-means, hierarchical clustering, and a Gaussian mixture model. The information processing method described in any of (1) to (3) above. (5) The regression model, when estimating the relationship between the skin function profile and the skin phenotypic information, includes one of the following groups as a confounding factor: sex, age, environmental factors, and lifestyle. The information processing method described in (2) or (3) above. (6) Based on the causal relationship estimated in the estimation step, a recommendation information generation step is further performed to generate recommendation information for improving the skin phenotype information. The information processing method described in any of (1) to (5) above. (7) The aforementioned recommended information includes information on at least one selected from the group consisting of skincare habits, lifestyle habits, skincare products, quasi-drugs, supplements, functional foods, and beauty devices. The information processing method described in (6) above. (8) An input unit that inputs at least three different types of skin function information and skin phenotype information, A profile creation unit that performs clustering using the aforementioned skin function information and creates a skin function profile, For each skin function profile, an occurrence index estimation unit estimates an occurrence index indicating the likelihood of the occurrence of the skin phenotypic information using a statistical model, An information processing apparatus comprising: an estimation unit that estimates the causal relationship between the skin function profile and the skin phenotype information using the statistical model, based on the occurrence index estimated by the occurrence index estimation unit and the skin function profile. (9) On the computer, An input step in which at least three different types of skin function information and skin phenotype information are entered, A profile creation step involves performing clustering using the aforementioned skin function information to create a skin function profile, For each skin function profile, an incidence index estimation step is performed to estimate an incidence index indicating the likelihood of the occurrence of the skin phenotypic information using a statistical model. A program for performing an estimation step, which involves estimating the causal relationship between the skin function profile and the skin phenotypic information using the statistical model, based on the incidence index estimated in the incidence index estimation step and the skin function profile. (10) A computer-readable recording medium on which the program described in (9) above is recorded. [Explanation of symbols]
[0102] 100 Information Processing Devices 101 Control Unit 102 Main storage 103 Auxiliary storage device 104 Input Interfaces 105 Output Interface 106 Communication Interface 107 Bus 201 Input section 202 Profile Creation Department 203 Occurrence Indicator Estimation Unit 204 Estimation section 205 Recommended Information Generation Unit S101 Input Step S102 Profile Creation Steps S103 Estimation step for occurrence indicator S104 Estimation Step S105 Recommended Information Generation Step
Claims
1. Computers An input step in which at least three different types of skin function information and skin phenotype information are entered, A profile creation step involves performing clustering using the aforementioned skin function information to create a skin function profile, For each skin function profile, an incidence index estimation step is performed to estimate an incidence index indicating the likelihood of the occurrence of the skin phenotypic information using a statistical model. An information processing method that performs an estimation step of estimating the causal relationship between the skin function profile and the skin phenotype information using the statistical model, based on the occurrence index estimated in the occurrence index estimation step and the skin function profile.
2. The estimation step involves using a regression model to estimate that the skin function profile causally contributes to the skin phenotypic information if the confidence interval of the regression coefficient corresponding to the variable representing the skin function profile does not contain zero at a predetermined confidence level. The information processing method according to claim 1.
3. The aforementioned confidence interval was calculated based on a 95% confidence level. The information processing method according to claim 2.
4. The clustering is performed using one of the following methods based on the input skin function information: k-means method, hierarchical clustering, and mixed Gaussian model. The information processing method according to any one of claims 1 to 3.
5. The regression model, when estimating the relationship between the skin function profile and the skin phenotypic information, includes one of the following groups as a confounding factor: sex, age, environmental factors, and lifestyle. The information processing method according to claim 2 or 3.
6. Based on the causal relationship estimated in the estimation step, a recommendation information generation step is further performed to generate recommendation information for improving the skin phenotype information. The information processing method according to claim 1.
7. The aforementioned recommended information includes information on at least one selected from the group consisting of skincare habits, lifestyle habits, skincare products, quasi-drugs, supplements, functional foods, and beauty devices. The information processing method according to claim 6.
8. An input unit that inputs at least three different types of skin function information and skin phenotype information, A profile creation unit that performs clustering using the aforementioned skin function information and creates a skin function profile, For each skin function profile, an occurrence index estimation unit estimates an occurrence index indicating the likelihood of the occurrence of the skin phenotypic information using a statistical model, An information processing apparatus comprising: an estimation unit that estimates the causal relationship between the skin function profile and the skin phenotype information using the statistical model, based on the occurrence index estimated by the occurrence index estimation unit and the skin function profile.
9. On the computer, An input step in which at least three different types of skin function information and skin phenotype information are entered, A profile creation step involves performing clustering using the aforementioned skin function information to create a skin function profile, For each skin function profile, an incidence index estimation step is performed to estimate an incidence index indicating the likelihood of the occurrence of the skin phenotypic information using a statistical model. A program for performing an estimation step, which involves estimating the causal relationship between the skin function profile and the skin phenotypic information using the statistical model, based on the incidence index estimated in the incidence index estimation step and the skin function profile.
10. A computer-readable recording medium on which the program described in claim 9 is recorded.
Citation Information
Patent Citations
Skin analysis method and skin analysis device
JP2018102730A
