JP7740949B2 - Food texture evaluation method - Google Patents

Description

The present invention relates to a method for evaluating the texture of food using a chewing behavior measuring device.

In recent years, consumer tastes in food have become increasingly diverse. A wide variety of food products are being developed daily to meet these diverse consumer preferences. Food development requires not only a product concept, but also a method for evaluating its distinctive texture and other characteristics.

A known method for measuring food texture is to use equipment similar to that used to measure the physical properties of general materials. For example, Patent Document 1 discloses a method that includes the steps of obtaining a tribology data set by measuring the friction factor of a food product as a function of sliding speed using a tribology device, and determining one or more sensory mouthfeel characteristics of the food product based on a previously determined correlation model.

Meanwhile, physiological measurements are known as a method for measuring the food intake behavior of humans (subjects). Physiological measurements involve evaluating the subject as if they were a measuring device, and examples include measuring myoelectric potential and swallowing sounds. While these physiological measurements are extremely useful for indirectly observing the subject's eating behavior, they are difficult to perform over long periods of time due to the complexity of the measurements and the burden they place on the subject, and there are also limits to the number of samples that can be measured at one time.

However, recently, a chewing behavior measuring device (bitescan, registered trademark, manufactured by Sharp Corporation) has been developed that can evaluate chewing behavior by simply measuring changes in jaw movement over time using a small, lightweight distance sensor attached to the ear. This device is an easy-to-use physiological measuring device that places little strain on the subject, making it possible to measure more natural chewing behavior over a longer period of time than conventional physiological measuring devices.

As an example of technology that uses such a chewing behavior measuring device, Patent Document 2 proposes a dietary monitoring method that uses a chewing behavior measuring device to measure the number of chews from changes in measurement values over time. It states that this makes it possible to estimate what is actually being eaten during a meal.

Patent No. 5548612 Japanese Patent Application Laid-Open No. 2020-58609

However, the food evaluation method described in Patent Document 1 evaluates texture from physical properties obtained through general physical property measurements, which means that there is an insufficient correlation between the evaluation results and the actual texture, and there are also problems with the types of texture that can be evaluated being limited.

Furthermore, the invention of Patent Document 2 uses a chewing behavior measuring device to simply measure the number of chews from changes in the measurement value over time, and therefore is not at a level where it can be used to evaluate the texture of food.

The object of the present invention is to provide a food texture evaluation method that uses a chewing behavior measuring device to accurately evaluate the texture of a wide range of foods.

The inventors discovered that food texture can be accurately evaluated by extracting appropriate parameters from chewing behavior data obtained using a chewing behavior measuring device and performing multiple regression analysis, and based on this finding, they have completed the present invention.

The present invention includes the following aspects:
Item 1.
a step of measuring, over time, detection values corresponding to jaw movement when the subject chews the target food using a chewing behavior measuring device;
a step of calculating a predicted value of the texture score of the target food by substituting the two or more parameters extracted from the waveform data of the detection values into a regression equation obtained in advance by multiple regression analysis using the texture scores of multiple foods as objective variables and the parameters as explanatory variables, and
A method for evaluating the texture of food comprising the steps of:

Item 2.
Item 2. A method for evaluating the texture of food according to Item 1, wherein the chewing behavior measuring device detects a distance corresponding to jaw movement and generates waveform data of the detected value.

Item 3.
Item 3. The method for evaluating the texture of food according to Item 1 or 2, wherein the parameters are two or more selected from the group consisting of a frequency parameter related to the chewing frequency, an area parameter related to the chewing peak area, a height parameter related to the chewing peak height, an area distribution parameter related to a histogram of the chewing peak area, a time parameter related to the chewing time, and a distance/time parameter related to a value obtained by dividing the chewing peak height by the duration of the chewing peak.

Item 4.
4. The method for evaluating the texture of food according to any one of Items 1 to 3, wherein the parameters are two or more selected from the group consisting of a total peak area ratio and an average peak area ratio for the early/late phases when the time until completion of mastication is divided into early, middle, and late phases, linear kurtosis and linear skewness in a histogram of the mastication peak areas, an average peak area, an average tempo value, a peak area of the first peak, the number of mastications, a maximum peak area, a time per mastication, a maximum peak height, and an average peak height.

Item 5.
5. The method for evaluating the texture of a food according to any one of Items 1 to 4, wherein the texture is one or more selected from the group consisting of hardness, elasticity, adhesion to teeth, adhesion to the tongue, ease of chewing, melt-in-the-mouth feel, degree of chewing (first half), degree of chewing (second half), and ease of holding together.

Item 6.
The multiple regression analysis is
a step of obtaining a score by quantifying the texture of five or more foods through a sensory evaluation in advance;
a step of measuring, over time, detection values corresponding to jaw movement when a subject chews the food using the chewing behavior measuring device;
a step of performing multiple regression analysis using the two or more parameters extracted from the waveform data of the detected values as explanatory variables and the food texture score as a response variable to obtain the regression equation;
Item 6. The method for evaluating the texture of food according to any one of Items 1 to 5, comprising:

Section 7.
The multiple regression analysis uses the scores and the parameters of a plurality of subjects,
Item 7. The method for evaluating the texture of foods according to Item 6, wherein, in order to reduce differences in the parameters among the plurality of subjects, the same reference value is set for any parameter of any food among the five or more foods for all subjects, and with this as a reference, relative values to the parameter of the reference food are used as the explanatory variables for the same parameters of other foods.

Section 8.
Item 8. The method for evaluating food texture according to any one of Items 1 to 7, further comprising: performing a preliminary multiple regression analysis in advance using the food texture score as a response variable and parameters extracted from the waveform data of the detection values as explanatory variables, in order to reduce the number of explanatory variables; and selecting a smaller number of explanatory variables than the explanatory variables used in the preliminary multiple regression analysis based on the standardized coefficients of the resulting regression equation.

Item 9.
Item 7. A method for evaluating the texture of foods according to Item 6, wherein the multiple regression analysis uses the two or more parameters as explanatory variables and the food texture score as a response variable to obtain the regression equation, and includes a step of creating a regression equation in which a number of explanatory variables smaller than the number of explanatory variables is selected by a stepwise method.

The present invention provides a food texture evaluation method that uses a chewing behavior measuring device to accurately evaluate the texture of a wide range of foods.

10 is a graph showing an example of waveform data when chewing gum. 10 is a graph showing an example of waveform data when chewing a soft candy. 10 is a graph showing an example of waveform data when chewing a gummy candy. 1 is a graph showing waveform data in which the time from the start of mastication to the end of mastication is divided into three equal parts: early, middle, and late periods. This is a histogram in which the frequency of each peak is counted for each certain division based on the chewing peak area. 1 is a graph showing the correlation between the actually measured values and the predicted values for the texture (ease of chewing) of an unknown food. 1 is a graph showing the correlation between the measured and predicted values for the texture (melt-in-the-mouth quality) of an unknown food.

(Method for evaluating food texture)
The food texture evaluation method of the present invention uses a chewing behavior measuring device to evaluate the texture of a target food. The target food may be any food whose texture can be evaluated, including existing foods as well as newly developed foods. Specific examples include the foods for which multiple regression analysis was performed in Examples 1 and 9-10, the foods for which predicted values were calculated as unknown foods in Example 11, and foods similar to these.

Foods that are subject to texture evaluation, including these, include frozen desserts such as ice cream, lacto ice cream, ice milk, and frozen desserts; gel-like and semi-solid foods such as jelly, pudding, and almond tofu; dairy products such as yogurt, cheese, and butter; confectioneries such as gummies, cookies, biscuits, snacks, rice crackers, chocolate, candy, cakes, and dumplings; beans such as soybeans, almonds, peanuts, and pistachios; dried foods such as dried squid, squid jerky, and nursing care foods such as dysphagia foods, chopped foods, and high-nutrition jellies; fruits such as mangoes, pears, muscat grapes, grapes, apples, bananas, melons, and peaches; carrots, onions, cucumbers, eggplants, cabbage, lettuce, green onions, and peaches. Examples of suitable foods include vegetables such as mashed potatoes, grains such as rice, wheat, barley, corn, soybeans, and red beans, meats such as beef, pork, chicken, goat, and lamb, and processed meat products such as sausages and ham, seafood such as tuna, sardines, saury, sea bream, salmon, sweetfish, shrimp, scallops, crab, and clams, and processed seafood products such as kamaboko and fish sausages, mushrooms such as shiitake and king oyster mushrooms, prepared dishes such as simmered dishes, fried dishes, stir-fried dishes, grilled dishes, and steamed dishes, noodles such as udon, soba, spaghetti, macaroni, and Chinese noodles (the noodles can be, for example, fresh noodles, semi-fresh noodles, frozen noodles, dried noodles, fried noodles, or non-fried noodles), breads such as white bread and whole wheat bread, and combinations of these foods.

The food texture evaluation method of the present invention includes a measurement step in which measurements are taken using a chewing behavior measuring device, and a calculation step in which a predicted value for the texture score of the target food is calculated. Each step is described below.

[Measurement process]
The method for evaluating the texture of food of the present invention includes a step of measuring, over time, detection values corresponding to jaw movement when a subject chews a target food using a chewing behavior measuring device. By measuring detection values corresponding to jaw movement over time, chewing behavior can be quantified into waveform data, making it possible to effectively extract multiple parameters for multiple regression analysis.

The number of subjects in the measurement may be one, but from the viewpoint of improving the accuracy of the evaluation using the average value, it is preferably three or more, more preferably five or more, and even more preferably seven or more. Furthermore, from the viewpoint of shortening the measurement time and simplifying the data processing, the fewer the number of subjects in the measurement, the better, and the upper limit of the number is preferably 15 or less, more preferably seven or less.
(Chewing behavior measuring device)
The chewing behavior measuring device may be any device that can measure detection values over time according to the jaw movement when the subject chews the target food, and examples of the detection values include a value that detects the distance between the sensor and the skin on the surface of the mandible, a value that detects the movement angle of the mandible, a value that detects a predetermined position of the mandible (for example, the lowest end), etc. These detection values can be measured without contacting the measurement site, so it is preferable to use a non-contact type chewing behavior measuring device.

An example of a non-contact type chewing behavior measuring device that can be used is the Bitescan (registered trademark) (product name BH-BS1RR) manufactured by Sharp Corporation. Chewing behavior measuring devices disclosed in JP 2016-131854 A, JP 2020-58609 A, etc. can also be used.

In particular, Bitescan uses a near-infrared distance sensor to detect the distance between the skin and the surface of the mandible in response to jaw movement, making it possible to measure the number of chews and chewing tempo. Real-time data can also be viewed on a smartphone app, and raw measurement data can be extracted as waveform data of detected values. A three-axis acceleration sensor also makes it possible to measure posture and movement. Because the device is compact, it places less strain on the subject during testing, allowing for long-term measurements. The ear-hook design minimizes the impact on the subject's eating behavior, making it possible to perform measurements in a manner similar to that of a normal meal, without any discomfort.

(waveform data)
Waveform data obtained by a chewing behavior measuring device can be, for example, data in which the horizontal axis represents time and the vertical axis represents the distance between the skin on the surface of the mandible and the sensor, as shown in Figures 1A to 1C. Figure 2A shows the relationship between the waveform data and the baseline, and in this example, the baseline is drawn on the upper side (the side with the larger distance). However, due to individual differences between subjects, some people have a larger distance when chewing and others have a smaller distance, so the baseline may be drawn on the lower side (the side with the smaller distance) for some subjects.

[Calculation process]
The method for evaluating the texture of foods of the present invention includes a step of using two or more parameters extracted from the waveform data of the detection values as explanatory variables, substituting the parameters into a regression equation obtained in advance by multiple regression analysis in which the texture scores of a plurality of foods are used as objective variables and the parameters are used as explanatory variables, to calculate a predicted value of the texture score of the target food.

(parameter)
The parameters extracted from the waveform data are preferably two or more selected from the group consisting of a frequency parameter (I) relating to the chewing frequency, an area parameter (II) relating to the chewing peak area, a height parameter (III) relating to the chewing peak height, an area distribution parameter (IV) relating to the histogram of the chewing peak area, a time parameter (V) relating to the chewing time, and a distance/time parameter (VI) relating to the value obtained by dividing the chewing peak height by the chewing peak duration. Although it is possible to select only one of these higher-level parameters (I) to (VI) and use multiple parameters of lower-level concepts in the multiple regression analysis, selecting two or more of the higher-level parameters (I) to (VI) enables more accurate evaluation of the texture of foods.

Frequency parameters (I) relating to chewing frequency include the average tempo, the average interval between peaks, the number of chews during the entire chewing time, and the number of chews within each time region when the chewing time is divided into three equal parts and labeled early, middle, and late in order from the closest to the chewing start point. Area parameters (II) relating to chewing peak area include the average peak area, the peak area of the first peak, the total peak area and average peak area within each time region when the time from the start of chewing to the end of chewing is divided into three equal parts and labeled early, middle, and late in order from the closest to the chewing start point. Examples of parameters include the total peak area ratio between the early and late phases, the average peak area ratio between the early and late phases, and the maximum peak area; height parameters (III) related to the mastication peak height include the average peak height, the peak height of the first peak, and the maximum peak height; area distribution parameters (IV) related to the histogram of the mastication peak area include the linear kurtosis and linear skewness, logarithmic kurtosis, and logarithmic skewness of the histogram of the mastication peak area; and time parameters (V) include the total mastication time from the start of mastication to the end of mastication, and the average time per mastication.

Furthermore, examples of the distance/time parameter (VI) related to the value obtained by dividing the chewing peak height by the chewing peak duration include the average value of the peak height/peak duration of each peak for the entire chewing time, and the average value of the peak height/peak duration within each time region when the chewing time is divided into three equal parts, which are labeled early, middle, and late, in order from the closest to the chewing start point.

More detailed definitions of these parameters include A to Rc below.
A: The total peak area ratio of the early/late period when the time from the start of chewing to the end of chewing is divided into three equal parts: early, middle, and late periods (total area of early peaks/total area of late peaks)
B: The average peak area ratio of the early/late period when the time from the start of chewing to the end of chewing is divided into three equal parts: early, middle, and late periods (average value of the early period peak area/average value of the late period peak area)
C: Linear kurtosis indicating the sharpness of the histogram of the mastication peak area (see FIG. 2B) (the kurtosis of the mastication peak area is calculated by the following formula (1)

D: Linear skewness indicating the asymmetry of the histogram of the chewing peak area (see Figure 2B) (skewness calculated for the chewing peak area using the following formula (2)).

E: Average peak area (average value of the peak areas of each peak from the start of chewing to the end of chewing)
F: Average tempo (number of chews per minute from the start of chewing to the end of chewing)
G: Peak area of the first peak (area of the peak after the first chewing)
H: Number of mastications from the start of mastication to the end of mastication I: Maximum peak area (maximum value of the peak area from the start of mastication to the end of mastication)
J: Time per chewing (average duration per peak from the start of chewing to the end of chewing)
K: Maximum peak height (maximum value of peak height from the start of chewing to the end of chewing)
L: Average peak height (average value of peak height from the start of chewing to the end of chewing)
Ma: Number of mastications in the early period when the time from the start of mastication to the completion of mastication is divided into thirds, namely, early, middle, and late periods. Mb: Number of mastications in the middle period when the time from the start of mastication to the completion of mastication is divided into thirds, namely, early, middle, and late periods. Mc: Number of mastications in the latter period when the time from the start of mastication to the completion of mastication is divided into thirds, namely, early, middle, and late periods. Na: Total peak area in the early period when the time from the start of mastication to the completion of mastication is divided into thirds, namely, early, middle, and late periods. Nb: Total peak area in the middle period when the time from the start of mastication to the completion of mastication is divided into thirds, namely, early, middle, and late periods. Peak area Nc: total peak area in the latter period when the time from the start of mastication to the completion of mastication is divided into thirds and divided into early, middle and late periods. Oa: average peak area in the early period when the time from the start of mastication to the completion of mastication is divided into thirds and divided into early, middle and late periods. Ob: average peak area in the middle period when the time from the start of mastication to the completion of mastication is divided into thirds and divided into early, middle and late periods. Oc: average peak area in the latter period when the time from the start of mastication to the completion of mastication is divided into thirds and divided into early, middle and late periods. Pa: average peak area in the latter period when the time from the start of mastication to the completion of mastication is divided into thirds and divided into early, middle and late periods. Average tempo value Pb in the early period when the time from the start of mastication to the completion of mastication is divided into thirds and divided into early, middle and late periods. Average tempo value Pc in the middle period when the time from the start of mastication to the completion of mastication is divided into thirds and divided into early, middle and late periods. Average tempo value Qa in the late period when the time from the start of mastication to the completion of mastication is divided into thirds and divided into early, middle and late periods. Average time per mastication in the early period Qb: Average time per mastication in the middle period when the time from the start of mastication to the completion of mastication is divided into thirds and divided into early, middle and late periods. The time from the start of mastication to the completion of mastication is divided into thirds, and the average value of the time per mastication in the latter period. Ra: The time from the start of mastication to the completion of mastication is divided into thirds, and the average value of the peak height/peak duration in the early period. Rb: The time from the start of mastication to the completion of mastication is divided into thirds, and the average value of the peak height/peak duration in the middle period. Rc: The time from the start of mastication to the completion of mastication is divided into thirds, and the average value of the peak height/peak duration in the latter period.

These parameters A to Rc can be categorized into higher order parameters (I) to (VI) as shown in Table 1A. Parameters Ma to Rc correspond to the early, middle, and late stages, and using these parameters individually can sometimes enable more accurate evaluation of food texture.

In the present invention, when multiple regression analysis is performed using the scores and parameters of multiple subjects, in order to reduce differences in the parameters between the multiple subjects, it is preferable to set the same reference value for one parameter of one of the multiple foods for all subjects, and use this as the reference value, and then use the relative values of the reference food parameter for the same parameter for other foods as the explanatory variables. This method not only reduces individual differences between subjects, but also reduces measurement error within individuals when measurements are taken on multiple days.

(Food texture)
The texture of the food to be evaluated may be any texture that a human (subject) feels while chewing, and may be one or more selected from the group consisting of hardness, elasticity, adhesion to teeth, adhesion to the tongue, ease of chewing, melt-in-the-mouth texture, chewiness (first half), chewiness (second half), chewiness, juiciness, freshness, melt-in-the-mouth texture, crunchiness, crunchy texture, firmness, crispiness, crispiness, crispness, crispiness, chewiness, firmness, elasticity, chewiness, fibrous texture, pulpy texture, stickiness, slimy texture, richness, smoothness, creamy texture, jiggly texture, popping texture, crispiness, fluffy texture, grainy texture, oily texture, addictive texture, crumbly texture, flaky texture, fluffy texture, granular texture, oily texture, addictive texture, crumbly texture, flaky texture, fluffy texture, and ease of holding together. The textures that can be evaluated with particular accuracy by the present invention using a chewing behavior measuring device include hardness, elasticity, adhesion to teeth, ease of chewing, melt-in-the-mouth texture, degree of chewing (first half), degree of chewing (second half), and ease of holding together.

When evaluating the texture of a food, criteria such as those shown in Table 2A in Example 1 and Table 22 in Example 9 can be used. These criteria can also be used when quantifying texture scores through sensory evaluation in multiple regression analysis. In other words, in the present invention, the texture of a food can be evaluated by quantifying the texture score through sensory evaluation and calculating a predicted texture score using the regression equation determined by multiple regression analysis.

Then, by selecting the top three parameters with the largest absolute values of the standardized coefficients in each example described below and listing all the top parameters to which the top three belong, the result is as shown in Table 1B. In other words, this table serves as an indicator of which top parameters should be selected as parameters to be used in multiple regression analysis in order to more accurately evaluate the texture of food, and it is preferable to select parameters from the top parameters in the table for each texture.

(Multiple regression analysis)
In multiple regression analysis, a regression equation is obtained in advance using the texture scores of multiple types of food as the objective variable and the parameters used in the calculation as the explanatory variables. The types of food to be subjected to multiple regression analysis are preferably 5 to 120 types, more preferably 10 to 80 types, and even more preferably 15 to 60 types.

The number of parameters used in multiple regression analysis should be two or more, but depending on the type of texture, a larger number of parameters may be preferable. Furthermore, from the perspective of shortening measurement time and simplifying data processing, the fewer the number of parameters, the better. Therefore, the number of parameters is preferably 2 to 12, and more preferably 2 to 7.

Specifically, the multiple regression analysis preferably includes the steps of: obtaining a score by quantifying the texture of a plurality of foods through a sensory evaluation; measuring, over time, detection values corresponding to jaw movement when the subject chews the food using the chewing behavior measuring device; and performing a multiple regression analysis using two or more parameters extracted from waveform data of the detection values as explanatory variables and the texture score as a response variable to obtain the regression equation.

The measurement of detected values for multiple regression analysis is preferably performed under the same conditions as the measurement process for evaluation, and the same subjects are also preferably used. However, by using waveform data from more subjects in the multiple regression analysis, it is possible to tolerate omissions or differences from some subjects, making it possible to perform food texture evaluation according to the present invention in an optimal manner.

Methods for quantifying texture scores through sensory evaluation include scoring methods using a numerical scale (e.g., 1 to 9 points), line scale methods using a linear scale, VAS (Visual Analog Scale), SD (Semantic Differential), QDA (Quantitative Descriptive Analysis), and TI (Time Intensity). Among these, the line scale method, which allows for continuous scoring according to degree, is preferred.

Multiple regression analysis can obtain not only the regression equation containing the partial regression coefficients of the explanatory variables, but also the standardized coefficients of the explanatory variables, probability values obtained from an analysis of variance of the regression and residual variations, the multiple correlation coefficient R of the regression equation, the adjusted R for degrees of freedom, and residuals indicating the difference between predicted and measured values. Furthermore, the validity of the regression equation can be determined from the probability values obtained from an analysis of variance of the regression and residual variations, and the accuracy of the regression equation can be determined from the adjusted R for degrees of freedom. The normality of the residuals can be evaluated using the Shapiro-Wilk test (SW test) or the Kolmogorov-Smirnov test (KS test).

By using these as indicators to select a combination of food texture and parameters to extract, it becomes possible to evaluate food texture with greater accuracy. Specifically, for example, combinations can be selected that result in an adjusted R of 0.5 or greater, a probability value of 0.05 or less obtained from an analysis of variance of regression variation and residual variation, and a SW test that indicates normality of residuals.

(regression equation)
Generally, in a regression equation (multiple regression equation) obtained by multiple regression analysis, the dependent variable is expressed as an equation in which an intercept is added to the sum of (partial regression coefficient × explanatory variable), i.e., dependent variable = Σ (partial regression coefficient × explanatory variable) + intercept.

Such a regression equation can be determined by performing multiple regression analysis using, for example, IBM SPSS Statistics (IBM), jmp (SAS), R (R Development Core Team), College Analysis (Fukuyama Heisei University), or the regression analysis function of Microsoft Excel (registered trademark) analysis tools.

In addition, by calculating the standardized coefficients from the partial regression coefficients of the regression equation, the contribution of each parameter can be determined from the magnitude of their absolute values.

(Calculation of predicted values)
By substituting two or more parameters, which are explanatory variables, into the regression equation obtained for the texture score, which is the objective variable, a predicted value for the texture score can be calculated, allowing the texture of the target food to be evaluated.

In other words, the accuracy of the evaluation of the calculated predicted score varies depending on the reliability of the regression equation used to calculate it (probability value and value of the multiple correlation coefficient adjusted for degrees of freedom). In this invention, by selecting a highly reliable combination of texture and parameters, it is possible to evaluate the texture of food with greater accuracy.

(Exploratory multiple regression analysis)
Furthermore, in the present invention, in order to reduce the number of explanatory variables, it is possible to perform a preliminary multiple regression analysis in advance using the food texture score as the response variable and parameters extracted from the waveform data of the detected values as explanatory variables, and then select a smaller number of explanatory variables than those used in the preliminary multiple regression analysis based on the standardized coefficients of the resulting regression equation.This makes it possible to maintain the accuracy of the food texture evaluation, shorten the measurement time, and simplify data processing.

The multiple regression analysis in exploratory multiple regression analysis can be carried out in the same way as the multiple regression analysis described above. The only difference is the number and type of parameters used. In other words, in the multiple regression analysis carried out after exploratory multiple regression analysis, only those parameters with the largest absolute values of the standardized coefficients in the regression equation are used as explanatory variables.

(Stepwise multiple regression analysis)
Instead of using preliminary multiple regression analysis as described above to select a smaller number of explanatory variables, it is also possible to perform multiple regression analysis using a stepwise method to create a regression equation with a smaller number of explanatory variables, and use this to substitute two or more parameter values (explanatory variables) to calculate a predicted value for the food texture score.

The statistical analysis software mentioned above (e.g., College Analysis ver. 6.7) has a stepwise method function that automatically extracts the optimal combination of explanatory variables. Using this function (stepwise method), it is possible to extract two to three explanatory variables for each texture. Stepwise methods include variable addition/deletion, variable addition/removal, and variable elimination, and variables are selected and determined based on numerical criteria such as the test probability (or test value) of the partial regression coefficient and the Akaike Information Criterion (AIC). At the same time, partial regression coefficients and intercepts are calculated, and a regression equation for calculating a predicted texture score can be obtained. Using multiple regression analysis with this stepwise method makes it possible to automatically extract more appropriate explanatory variables, reduce the number of parameters extracted from waveform data, and more efficiently evaluate food texture.

The present invention will be explained in detail below using examples, but the scope of the present invention is not limited to these examples.

Example 1
(Chewing behavior measuring device)
The chewing behavior measurement device used was the Bitescan (registered trademark) (product name BH-BS1RR) manufactured by Sharp Corporation. This device uses a near-infrared distance sensor to detect the distance between the skin and the surface of the mandible in response to jaw movement, making it possible to measure the number of chews and chewing tempo. Real-time data can also be viewed using a smartphone app, and raw measurement data can be extracted as waveform data of detected values. A three-axis acceleration sensor can also be used to measure posture and movement, and although this function was not used in this study, it is also possible to use parameters extracted from the results of measurements using this sensor.

(waveform data)
Examples of waveform data obtained by the chewing behavior measuring device are shown in Figures 1A to 1C. Figure 1A is an example of waveform data obtained when chewing a commercially available gum, Figure 1B is an example of waveform data obtained when chewing a commercially available soft candy, and Figure 1C is an example of waveform data obtained when chewing a commercially available gummy candy. In these figures, one peak in the waveform corresponds to one chew, and it can be seen that multiple peaks in the waveform correspond to the number of chews, and exist at a fairly constant tempo. It can also be seen that the peak area of each waveform changes during chewing.

As shown in Figures 1A and 1B, chewing gum and soft candy is done at a constant tempo and involves a large number of chews, whereas as shown in Figure 1C, chewing of gummy candy is irregular and involves fewer chews. Therefore, it can be said that Bitescan can measure differences in chewing behavior when eating foods with different textures.

(Parameter extraction)
To extract the parameters, the time from the start of mastication to the end of mastication in the waveform data was divided into three equal parts: early, middle, and late periods, as shown in Figure 2A. Furthermore, as shown in Figure 2B, the frequency of each peak was counted for each division based on the mastication peak area to create a histogram. Based on these, parameters to be used as explanatory variables in multiple regression analysis were extracted from the waveform data obtained by measurement. Specifically, the following parameters A to G were extracted.
A: The total peak area ratio of the early/late period when the time from the start of chewing to the end of chewing is divided into three equal parts: early, middle, and late periods (total area of early peaks/total area of late peaks)
B: The average peak area ratio of the early/late period when the time from the start of chewing to the end of chewing is divided into three equal parts: early, middle, and late periods (average value of the early period peak area/average value of the late period peak area)
C: Linear kurtosis calculated from the distribution of mastication peak areas using the formula (1) D: Linear skewness calculated from the distribution of mastication peak areas using the formula (2) E: Average peak area (average value of the peak areas of each peak from the start of mastication to the end of mastication)
F: Average tempo (number of chews per minute from the start of chewing to the end of chewing)
G: Peak area of the first peak (area of the peak after the first chewing)
In order to reduce differences between subjects, the values of E, F, and G above were calculated as parameters for the other foods relative to the value obtained when all subjects chewed the gum 30 times, which was set to 1. However, for G, the value obtained by dividing the value by the average peak area of E for the gum was calculated as a parameter for all foods.

(Sensory evaluation of texture)
The 19 foods listed in Table 2B were subjected to sensory evaluation of the texture shown in Table 2A using the line scale method, and texture scores were determined as the objective variable for multiple regression analysis. Specifically, the right end of a 100 mm line was defined as the maximum and the left end as the minimum. Based on the definitions, subjects marked any location on the 8 texture items listed in Table 2A, and the distance from the left end was used as the sensory evaluation score (texture score). The subjects consisted of seven trained panelists (five men and two women in their 20s to 40s). The 19 foods listed in Table 2B (weight, shape, and size as shown in Table 2C) were used as target foods. The food intake was ad libitum. Texture terms and definitions were determined through discussion and discussion among the subjects. The evaluation scale was also agreed upon among the subjects before the evaluation.

(Multiple regression analysis)
A multiple regression analysis was performed using statistical analysis software College Analysis ver. 6.7, with the scores obtained from the sensory evaluation of the 19 foods as the objective variable and parameters A to G extracted from the waveform data measured for the 19 foods (weight, shape, and size are as shown in Table 2C) as the explanatory variables. Among the results, for the cases where the objective variable was the score for "ease of chewing," the values of each explanatory variable for each food are shown in Table 2B. Here, the objective variable and explanatory variables are the average values for seven subjects.

As a result of multiple regression analysis, the following regression equation was obtained from the partial regression coefficients.
Ease of chewing = 49.5A + 9.1B - 17.6C + 47.4D - 298.4E - 190.5F + 66.6G + 383.6
The standardized coefficients obtained at that time are shown in Table 3 together with the partial regression coefficients.

The results in Table 3 show that the contributions of C, D, and E are relatively large (in multiple regression analysis, the larger the absolute value of the standardized coefficient, the greater the contribution). Furthermore, the multiple correlation coefficient R obtained by multiple regression analysis was 0.838, and the adjusted R was 0.715. Here, since the variation expressed by the regression increases as the degrees of freedom increase, the adjusted coefficient of determination ^R2 is obtained by adjusting for the degree of freedom, and its square root is the adjusted R.

For textures other than "ease of chewing," regression equations were also obtained from the partial regression coefficients obtained, and the normality of the residuals for the regression equation, probability value, validity of the regression equation, multiple correlation coefficient R, and adjusted R are shown in Table 4A. The results of the sensory evaluation of the 19 foods are shown in Table 4B.

As the results in Table 4A show, the probability values for elasticity, ease of chewing, melt-in-the-mouth texture, chewiness (first half), and chewiness (second half) were small, confirming the validity of the regression equation, with adjusted R values of 0.68 or higher. Here, the probability values were determined by analyzing the variance ratio between the regression variation and the residual variation (same below), and the validity of the regression equation was marked as 'O' when the probability value was 0.05 or less, and 'X' when it exceeded 0.05 (same below). In this way, for textures where the validity of the regression equation was confirmed, the texture of food can be accurately evaluated (predicted) without conducting a sensory evaluation.

Using the parameters averaged over seven people, we investigated which parameters had the greatest contribution to the eight texture characteristics. The results are shown in Table 5.

As the results in Table 5 show, the parameters with the greatest contribution vary depending on the texture. In multiple regression analysis, the larger the absolute value of the standardized coefficient, the greater the contribution of the parameter.

Example 2 (Subject a only)
(multiple regression analysis)
In Example 1, measurements were performed using the chewing behavior measuring device only on subject a, and parameters were extracted from the obtained waveform data, but multiple regression analysis was performed using the same method as in Example 1. Among the results, when the objective variable was the score of "ease of chewing," the values of each explanatory variable for each food are shown in Table 6.

The regression equation obtained from the partial regression coefficients was as follows:
Ease of biting = -1.1A - 89.6B - 25.4C + 56.8D - 313.4E - 16.4F - 7.1G + 447.8
The standardized coefficients obtained at that time are shown in Table 7 together with the partial regression coefficients.

The results in Table 7 show that, as with the case where the average values of seven people were used, the contributions of C, D, and E were relatively large. Furthermore, the multiple correlation coefficient R from the multiple regression analysis was 0.818, and the adjusted R was 0.678. Furthermore, the probability value for the regression equation was 0.042, confirming the validity of the regression equation.

Therefore, even when multiple regression analysis is performed using measurements from a single subject using a chewing behavior measurement device, food texture can be evaluated with high accuracy for textures for which the regression equation is found to be valid.

Example 3 (Subject b only)
(Multiple regression analysis)
In Example 1, measurement was performed using the chewing behavior measuring device only on subject b, and parameters were extracted from the obtained waveform data, but multiple regression analysis was performed using the same method as in Example 1. Among the results, when the objective variable was the score of "ease of chewing," the values of each explanatory variable for each food are shown in Table 8.

The regression equation obtained from the partial regression coefficients was as follows:
Ease of chewing = 37.6A + 81.9B + 6.2C - 12.5D + 34E - 9.7F - 65.3G - 66.1
The standardized coefficients obtained at that time are shown in Table 9 together with the partial regression coefficients.

The results in Table 9 show that, unlike when the average value of seven people was used, the contributions of A, B, and G were relatively large. Furthermore, the multiple correlation coefficient R from the multiple regression analysis was 0.902, and the adjusted R was 0.833.

For textures other than "ease of chewing," regression equations were also obtained from the partial regression coefficients obtained, and the normality of the residuals for the regression equation, probability value, validity of the regression equation, multiple correlation coefficient R, and adjusted R are shown in Table 10, along with the case where the average values of seven subjects were used (Example 1).

As the results in Table 10 show, for subject b, the probability values for hardness, elasticity, adhesion to teeth, ease of chewing, melt-in-the-mouth texture, chewing intensity (first half), and chewing intensity (second half) were small, confirming the validity of the regression equation, and the adjusted R was 0.66 or greater. In this way, even without using the average values of seven people, for textures where the regression equation is found to be valid, it is possible to accurately evaluate (predict) the texture of food without conducting a sensory evaluation.

Comparative Example 1 (Evaluation by Physical Property Measurement)
The breaking stress and breaking strain of the 19 food samples used in Example 1 were measured using a texture analyzer (product number TA-XT2i, manufactured by Stable Micro Systems), and the correlation with the sensory evaluation score was confirmed. The measurement conditions (indentation test, penetration) were a probe diameter of 3 mm, a compression speed of 10 mm/s, and a measurement temperature of 20°C. The correlation was confirmed by simple regression analysis using the breaking stress or breaking strain as an explanatory variable.

The correlation coefficients between the resulting texture scores and the physical property values are shown in Table 11.

As the results in Table 11 show, the correlation between the sensory evaluation and physical property values in Comparative Example 1 was low compared to Examples 1 to 3. This demonstrates that the food texture evaluation method of the present invention is more accurate than texture evaluation methods that rely on physical property measurements.

Example 4 (when the number of parameters is 2)
Multiple regression analysis was performed in the same manner as in Example 1, except that the number of parameters was limited to two in descending order of contribution (largest absolute value of standardized coefficient) based on Table 5. The normality of the residuals for the resulting regression equation, probability value, validity of the regression equation, multiple correlation coefficient R, and adjusted R are shown in Table 12, along with the case where the number of parameters is 7 (Example 1).

As the results in Table 12 show, when the number of parameters is two, the effectiveness of the regression equation is confirmed for hardness, adhesion to teeth, and adhesion to the tongue, and it is clear that food texture can be accurately evaluated not only when the number of parameters is seven (Example 1), but also when the number of parameters is two. Here, the normality of the residuals was determined to be normal using an SW test with a significance level of 0.05, and was marked with a 'O' when it was determined to be non-normal (same below). In particular, it can be seen that by performing a multiple regression analysis using a larger number of parameters in advance and selecting from those parameters the parameters with the greatest contribution, it is possible to accurately evaluate food texture even with fewer parameters.

Example 5 (when the number of parameters is 3)
Multiple regression analysis was performed in the same manner as in Example 1, except that the number of parameters was limited to three in descending order of contribution (largest absolute value of standardized coefficient) based on Table 5. The normality of the residuals for the resulting regression equation, probability value, validity of the regression equation, multiple correlation coefficient R, and adjusted R are shown in Table 13, along with the case where the number of parameters is 7 (Example 1).

As the results in Table 13 show, when the number of parameters is three, the effectiveness of the regression equations is confirmed for hardness, adhesion to teeth, and adhesion to the tongue, and it is clear that food texture can be accurately evaluated not only when the number of parameters is seven (Example 1), but also when the number of parameters is three. In particular, it is clear that by performing multiple regression analysis using a larger number of parameters in advance and selecting from those parameters the parameters with the greatest contribution, food texture can be accurately evaluated even with fewer parameters.

Example 6 (number of parameters: 5)
In Example 1, multiple regression analysis was performed using the same method as in Example 1, except that the parameters were limited to five in descending order of contribution (largest absolute value of standardized coefficient) based on Table 5. In this case, only "ease of chewing," for which the effectiveness of the regression equation was low when the number of parameters was 2 or 3, was the subject of multiple regression analysis. The normality of the residuals for the resulting regression equation, probability value, effectiveness of the regression equation, multiple correlation coefficient R, and adjusted R are shown in Table 14.

As the results in Table 14 show, when the number of parameters is 5, the probability value for the regression equation for "ease of chewing" is lower than when the number of parameters is 2 or 3 (Examples 4 and 5), confirming the validity of the regression equation and indicating that there are textures for which increasing the number of parameters is effective. For such textures, increasing the number of parameters makes it possible to accurately evaluate the texture of the food.

Comparative Example 2 (Simple Regression Analysis)
Regression analysis was carried out under the same conditions as in Example 1, except that instead of performing multiple regression analysis in Example 1, a regression equation was created by performing simple regression analysis using each of parameters A to G as explanatory variables, and the correlation coefficient was calculated. The resulting correlation coefficients for each parameter with respect to the texture score are shown in Table 15, along with the adjusted R obtained in Example 1.

As the results in Table 15 show, the correlation between the sensory evaluation and each parameter in Comparative Example 2 is low (52 out of a total of 56 combinations) compared to the adjusted R in Example 1. This shows that the food texture evaluation method of the present invention is significantly more accurate than the texture evaluation method using simple regression analysis.

Example 7 (using parameters H to L)
Multiple regression analysis was performed in the same manner as in Example 1, except that instead of using parameters A to G, the following parameters H to L were used. Table 16 shows the values of each parameter (average of seven subjects) extracted for each food (weight, shape, and size are as shown in Table 2C). Table 17 shows the standardized coefficients for the texture scores of each parameter obtained as a result, and Table 18 shows the normality of the residuals for the regression equation, probability value, validity of the regression equation, multiple correlation coefficient R, and adjusted R.
H: Number of mastications from the start of mastication to the end of mastication I: Maximum peak area (maximum value of the peak area from the start of mastication to the end of mastication)
J: Time per chewing (average duration per peak from the start of chewing to the end of chewing)
K: Maximum peak height (maximum value of peak height from the start of chewing to the end of chewing)
L: Average peak height (average value of peak height from the start of chewing to the end of chewing)
In order to reduce differences between subjects, the values of I, J, K, and L above were calculated as parameters for other foods relative to the value of gum set to 1 for all subjects.

As the results in Table 18 show, compared to when parameters A to G were used (Example 1), the probability values for the regression equation for hardness and adhesion to teeth were lower, confirming the validity of the regression equation. It can also be seen that the adjusted R for adhesion to the tongue was larger.

Example 8 (using parameters A to L)
Multiple regression analysis was performed in the same manner as in Example 1, except that instead of using parameters A to G in Example 1, parameters A to L were added to parameters H to L. The values of each parameter A to G (average values of seven subjects) extracted for each food are shown in Table 2B. Table 19 shows the standardized coefficients for the texture scores of each parameter obtained as a result, and Table 20 shows the normality of the residuals for the regression equation, probability values, validity of the regression equation, multiple correlation coefficient R, and adjusted R.

As the results in Table 20 show, comparable results were obtained compared to when parameters A to G were used (Example 1).

Comparative Example 3 (Simple Regression Analysis)
Regression analysis was carried out under the same conditions as in Example 7, except that instead of performing multiple regression analysis in Example 7, a regression equation was created by performing simple regression analysis using each of the parameters H to L as explanatory variables, and the correlation coefficient was calculated. The resulting correlation coefficients for each parameter with respect to the texture score are shown in Table 21, along with the adjusted R obtained in Example 7.

As the results in Table 21 show, the correlation between the sensory evaluation and each parameter in Comparative Example 3 is low (40 out of a total of 40 combinations is low) compared to the adjusted R in Example 7. This shows that the food texture evaluation method of the present invention is significantly more accurate than the texture evaluation method using simple regression analysis.

Example 9 (Texture evaluation of food crushed with the tongue)
In Example 1, except for the following changes, food that can be eaten by crushing with the tongue, as shown in Table 23, was subjected to a sensory evaluation in advance for texture (adhesion to the tongue, melt-in-the-mouth texture, and ease of collection), and then Bitescan measurement was performed. A multiple regression analysis was performed using the food texture score as the dependent variable and the extracted parameters as the explanatory variables to create a regression equation, and its effectiveness was examined.

For the Bitescan measurements, four subjects (three men and one woman) in their 20s and 30s were asked to cut out a 10g±0.5g sample from each food item and crush it with their tongues 15 times before eating it. For the sensory evaluation, the same four subjects (three men and one woman) in their 20s and 30s as above were asked to evaluate the texture using the same line scale method as in Example 1, based on the texture definitions in Table 22. Table 23 also shows the texture scores for each food item, which was the objective variable.

Table 24 shows parameters A to G (explanatory variables) used in the multiple regression analysis, and the method for extracting parameters A to G is the same as in Example 1.

In this example, after performing a preliminary multiple regression analysis (creating a regression equation using all parameters A to G) as in Example 1, three or four highly contributing parameters, as shown in Table 25, were selected and a further multiple regression analysis was performed. Then, a regression equation was created including the obtained partial regression coefficients and intercepts, and the validity of the regression equation was examined. The results are shown in Table 26.

As shown in Table 26, for all textures, the probability values were small, confirming the validity of the regression equation, and the adjusted R was 0.90 or higher. Thus, for textures for which the regression equation is found to be valid, the texture of the food can be accurately evaluated (predicted) without performing a sensory evaluation. Therefore, it was confirmed that the texture of food can be accurately evaluated even for foods that are crushed with the tongue.

Example 10 (when parameters Ma to Rc are used)
In Example 1, the foods and textures used to create the equations are those shown in Table 27, and the same subject (a man in his 30s) was used for both the Bitescan measurements and the sensory evaluation. Multiple regression analysis was performed using the same method as in Example 1, except that multiple regression analysis was performed for each texture using the variable addition/decrement method using parameters Ma to Rc. Table 27 also shows the texture scores of each food, which are the objective variables. Table 28 shows an overview of parameters Ma to Rc, and Table 29 shows the values of each parameter Ma to Rc extracted for each food. Note that Na to Rc were calculated as relative values when the value obtained after chewing the gum 30 times was set to 1.

The statistical analysis software College Analysis ver. 6.7 has a variable addition/deletion function that automatically extracts the optimal combination of explanatory variables. By using this function (variable addition/deletion), two or three explanatory variables were extracted for each texture. These are shown in Table 30, along with the partial regression coefficients and standardized coefficients.

From the results, the following regression equations (prediction equations) were obtained for each texture.
Ease of chewing = -68.1Na + 94.3Ra - 14.9
Melt-in-mouth quality = 81.7Ra - 1.6Mc - 125.3Rc + 82.9
The results of examining the validity of these regression equations are shown in Table 31.

As shown in Table 31, for all textures, the probability values were small, the validity of the regression equation was confirmed, and the adjusted R values were also high.

Example 11 (Prediction of unknown foods using the regression equation of Example 10)
Those skilled in the art will recognize that the effectiveness of the regression equation of Example 10 means that it can be used to accurately evaluate the texture of foods. To confirm this, texture predictions were made for the unknown samples shown in Table 32, and the differences from the actual measured values were confirmed. Specifically, Bitescan measurements were performed using the same method as in Example 10, and the parameters Na, Ra, Mc, and Rc were extracted and substituted into the regression equation of Example 10 to calculate predicted texture scores for the unknown foods. The results of extracting the parameters Na, Ra, Mc, and Rc are shown in Table 33. The relationship between the actual measured values and the predicted values is also shown in Figures 3A and 3B.

As shown in Figures 3A and 3B, there was a high correlation between the measured and predicted values for the texture of the unknown food (ease of chewing, melt-in-the-mouth quality), confirming the effectiveness of the present invention, as predicted by the validity of the regression equation.

Claims (9)

a step of measuring detection values corresponding to jaw movement over time when a subject chews a target food using a chewing behavior measuring device equipped with a distance sensor ;
a step of calculating a predicted value of the texture score of the target food by substituting the two or more parameters extracted from the waveform data of the detection values into a regression equation obtained in advance by multiple regression analysis using the texture scores of multiple foods as objective variables and the parameters as explanatory variables, and
A method for evaluating the texture of food comprising the steps of: The food texture evaluation method according to claim 1, wherein the chewing behavior measuring device detects a distance corresponding to jaw movement and generates waveform data of the detected value. The method for evaluating the texture of food according to claim 1 or 2, wherein the parameters are two or more selected from the group consisting of a frequency parameter relating to the chewing frequency, an area parameter relating to the chewing peak area, a height parameter relating to the chewing peak height, an area distribution parameter relating to a histogram of the chewing peak area, a time parameter relating to the chewing time, and a distance/time parameter relating to the value obtained by dividing the chewing peak height by the duration of the chewing peak. 4. The method for evaluating the texture of food according to any one of claims 1 to 3, wherein the parameters are two or more selected from the group consisting of: a total peak area ratio and an average peak area ratio for the early/late phases when the time until the completion of mastication is divided into early, middle, and late phases; linear kurtosis and linear skewness in a histogram of mastication peak areas; an average peak area; an average tempo value; a peak area of the first peak; the number of mastications; a maximum peak area; a time per mastication; a maximum peak height; and an average peak height. The method for evaluating the texture of food according to any one of claims 1 to 4, wherein the texture is one or more selected from the group consisting of hardness, elasticity, adhesion to the teeth, adhesion to the tongue, ease of chewing, melt-in-the-mouth texture, chewiness (first half), chewiness (second half), and ease of assembly. The multiple regression analysis is
a step of obtaining a score by quantifying the texture of five or more foods through a sensory evaluation in advance;
a step of measuring, over time, detection values corresponding to jaw movement when a subject chews the food using the chewing behavior measuring device;
a step of performing multiple regression analysis using the two or more parameters extracted from the waveform data of the detected values as explanatory variables and the food texture score as a response variable to obtain the regression equation;
The method for evaluating the texture of food according to any one of claims 1 to 5, comprising: The multiple regression analysis uses the scores and the parameters of a plurality of subjects,
7. The method for evaluating the texture of foods according to claim 6, wherein, in order to reduce differences in the parameters among the plurality of subjects, the same reference value is set for any parameter of any food among the five or more foods for all subjects, and with this as a reference, for the same parameter of other foods, relative values with respect to the parameter of the reference food are used as the explanatory variables. The food texture evaluation method of any one of claims 1 to 7 further comprises: performing a preliminary multiple regression analysis in advance, using the texture score as the objective variable and parameters extracted from the waveform data of the detection values as explanatory variables, in order to reduce the number of explanatory variables; and selecting a smaller number of explanatory variables than those used in the preliminary multiple regression analysis based on the standardized coefficients of the resulting regression equation. The food texture evaluation method according to claim 6, wherein the multiple regression analysis uses the two or more parameters as explanatory variables and the texture score as a response variable to obtain the regression equation, and includes a step of creating a regression equation in which a number of explanatory variables less than the number of explanatory variables is selected by a stepwise method.