Digestion evaluation method

Abstract

To provide a method for quantitatively evaluating the progress degree of digestion of food.SOLUTION: A stomach body constituting the stomach model device is filled with artificial digestive juice, food is put therein, the stomach model is made to perform a movement similar to the peristaltic movement of the stomach, and a color moving image of the digestive state of the food with time in a specific range is photographed. Then, texture analysis (average value of luminance, variance of luminance, contrast of luminance, skewness of luminance, kurtosis of luminance, energy of luminance, entropy of luminance) is performed on the captured image to quantify the digestion state.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method for quantitatively evaluating digestive behavior by analyzing texture of images from a digestion test using an in vitro gastric digestion model device. [Background technology]

[0002] In the stomach, in addition to chemical digestion by digestive enzymes, physical digestion also occurs, such as breaking food into small particles by the squeezing movement of the stomach wall called peristalsis. The physical characteristics (texture) of food in the body are important, as can be seen from the fact that the theme of the first World Food Summit was "Measurement and Perception of Texture." Furthermore, quantitative evaluation of food digestion in the stomach will lead to the development of foods with controlled gastric digestion behavior.

[0003] Patent Document 1 proposes an in vitro stomach digestion model device (stomach model device) that performs movements similar to the peristaltic movements of the human stomach and that can be observed from the outside.

[0004] Patent Documents 2 and 3 describe the use of image analysis including texture analysis in image diagnosis (ultrasound echo, MRI, etc.) in medical settings.

[0005] Patent documents 4 and 5 and non-patent documents 8 and 9 describe studies on extracting values ​​related to the chemical digestion of food based on images of food samples immersed in artificial digestive fluids.

[0006] Non-Patent Document 1 describes the standard operating procedure for the stomach model device described in Patent Document 1.

[0007] Non-Patent Document 2 states that the texture (physical properties) of food, along with its taste and aroma, is a factor that determines the palatability of food, and lists hardness, adhesiveness, and cohesiveness as parameters that can be determined by TPA (Texture Profile Analysis) measurements.

[0008] Non-Patent Document 3 explains that texture is the physical characteristic of food that is perceived by humans, and in a broad sense, it includes properties perceived visually and aurally, and in a narrow sense, it refers to mechanical properties perceived while eating, such as feel on the tongue, crispness, chewiness, and smoothness down the throat. Note that texture in the present invention refers to properties perceived visually.

[0009] Non-Patent Documents 4 and 5 report research into oral digestion, in which texture analysis is performed on images of a food bolus (a mass of food that has been chewed and mixed with saliva) that has been chewed and then spat out by a subject.

[0010] Non-Patent Documents 6, 7, and 8 report clinical studies in which the behavior of the stomach and its contents is visualized using magnetic resonance imaging (MRI) and the texture features of the obtained images are analyzed (NMR).

[0011] Non-patent documents 9 and 10 report studies in which food samples were immersed in artificial digestive fluids in an in vitro gastric digestion test, and images were taken using hyperspectral imaging, NMR, or MRI, and information on digestive behavior was obtained through image analysis.

[0012] Furthermore, Non-Patent Documents 11 to 14 are documents relating to the use of machine learning in the field of digestion. Non-patent literature 11 and 12 are examples of machine learning being used in the diagnostic imaging of digestive diseases in the medical field. Non-patent literature 13 is an example of machine learning prediction of the mechanical properties of an apple from micro-CT image data collected during ex vivo gastric digestion of an apple. Non-patent literature 14 is an example of machine learning prediction of satiety based on the viscosity of gastric contents during ex vivo gastric digestion. [Prior art documents] [Patent documents]

[0013] [Patent Document 1] Japanese Patent Application Laid-Open No. 2014-142535 [Patent Document 2] Japanese Patent Publication No. 2020-130596 [Patent Document 3] Japanese Patent Publication No. 2022-1841140 [Patent Document 4] Japanese Patent Publication No. 2022-90414 [Patent Document 5] Japanese Patent Application Laid-Open No. 2012-254036 [Non-patent literature]

[0014] [Non-Patent Document 1] NARO, Human Gastric Digestion Simulator Standard Operating Procedures, Public Version, October 27, 2022, https: / / www.naro.go.jp / publicity_report / publication / laboratory / naro / sop / 156536.html [Non-patent document 2] "Physical Properties of Food and Issues in Measurement" by Katsuyoshi Nishinari, Journal of the Japanese Society of Home Economics Vol. 64 No. 12 811-822 (2013) [Non-patent document 3] "Evaluation of Taste by Texture Analysis" Kaoru Kamiyama Chemistry and Biology Vol.47.No.2.2009 [Non-patent document 4] Tournier, C., Grass, M., Zope, D., Salles, C., & Bertrand, D. (2012). Characterization of bread breakdown during mastication by image texture analysis. Journal of Food Engineering, 113(4), 615-622. https: / / doi.org / https: / / doi.org / 10.1016 / j.jfoodeng.2012.07.015 [Non-Patent Document 5] Morell , P. , Hernando , I. , & Fiszman , S. (2014). Understanding the relevance of in-mouth food processing. A review of in vitro techniques. Trends in Food Science & Technology, 35(1), 18-31. https: / / doi.org / https: / / doi.org / 10.1016 / j.tifs.2013.10.005

Outdoor Configuration6

Direct Environment 7

Outdoor Track 8

[0015] Regarding oral digestion, subjects are asked to chew food and then spit it out, and texture analysis is performed on images of the resulting food bolus (food that has been chewed and mixed with saliva). However, there is insufficient technology to analyze the digestive behavior in the stomach and small intestine from images after food has been chewed and swallowed. While MRI imaging of stomach and small intestinal contents and image analysis to obtain information on digestive behavior are expensive, clinical trials pose ethical issues, and MRI visualizes the distribution of water and oil in grayscale, which provides limited information compared to color images obtained using visible light. In in vitro digestion tests that do not involve clinical trials, the subject is imaged using NMR or MRI, and information on digestive behavior is obtained through image analysis. This method simulates chemical digestion by immersing the sample in an artificial digestive fluid, and does not take into account the physical digestion of food through gastric peristalsis. Furthermore, when using a stomach model device that simulates gastric peristalsis, it is difficult to expose it to the magnetic field of NMR or MRI because it contains electronic components and metal materials.

[0016] Even in cases where machine learning has been used, it is not possible to numerically determine the degree of digestibility from external characteristics (texture). [Means for solving the problem]

[0017] In order to solve the above problems, the digestion evaluation method of the present invention performs texture analysis using color images of food obtained from a stomach model device, and quantifies the degree of digestion based on temporal changes in at least one of the texture features calculated from the mean brightness, brightness variance, brightness contrast, brightness skewness, brightness kurtosis, brightness energy, and brightness entropy.

[0018] The texture analysis can be performed after converting a color image to grayscale, or by using a color channel depending on the type of food. For example, the red channel can be used for analyzing the texture of carrots, and the green channel can be used for analyzing the texture of spinach.

[0019] In addition, the digestion evaluation method based on machine learning according to the present invention captures the brightness information (brightness intensity and position information) of the color image of the food obtained from the stomach model device as texture features, and has a computer learn the brightness information at the start and end of the digestion test as the undigested state and the digested state, respectively, to create a learning model, which is then used to quantify the degree of digestion in the image to be measured. [Effects of the Invention]

[0020] By subjecting videos or images obtained from an externally observable stomach model device to texture analysis, digestive behavior can be quantified directly from the videos or images without experimental analysis. In other words, it is possible to obtain information on gastric digestion that takes into account the physical digestion of food through gastric peristalsis, making it possible to shorten the time required to develop new foods.

[0021] Since time series information for each texture feature can be obtained, temporal changes in digestive behavior can be easily understood from the images.

[0022] When color images are used as image information, the amount of information is greater than that of grayscale images, and the degree of freedom in analysis is increased. For example, if the food is red, such as a carrot, the red channel is selected, and if the food is green, such as leafy vegetables, the green channel is selected. The appropriate color channel can be selected depending on the food sample.

[0023] It is possible to obtain texture features with high temporal resolution according to the frame rate of the video, and by curve fitting the time change plot, it is possible to quantify the degree of progress of gastric digestion using a rate constant.

[0024] By applying preprocessing (filtering) to the image to be analyzed to emphasize its features, differences in texture features can be emphasized, making comparison easier.

[0025] The quantitative assessment of digestion according to the present invention is entirely carried out in vitro, and therefore does not raise ethical issues that arise in clinical trials.

[0026] Furthermore, when machine learning is applied to the digestion evaluation method, once a learning model is created, the degree of digestion can be known simply by inputting the image for which digestion is to be evaluated into the learning model without requiring any special skills or knowledge, thereby expanding the range of applications. [Brief explanation of the drawings]

[0027] [Figure 1] 1(a) to 1(c) are diagrams illustrating an outline of the method of the present invention. [Figure 2] (a) to (c) are diagrams showing examples of the image to be analyzed, the brightness histogram obtained from the image, and the texture features obtained from the histogram analysis. The images shown are test images, which are different from the images obtained by the stomach model device. [Figure 3] These are photographs taken at various times during a digestion test of mashed potato gel using a stomach model device: (a) is taken at 0 min, (b) is taken approximately 10 min, and (c) is taken approximately 180 min after the start. [Figure 4] This graph shows the time change in texture features in the grayscale image of the color image during the digestion test shown in Figure 3. (a) is the change in variance of brightness, (b) is the change in energy of brightness, and (c) is the change in entropy of brightness. Each plot data shows a value calculated at 30-second intervals. [Figure 5] This graph shows the results of the same test as in Figure 4, but with only the red channel extracted. [Figure 6] These graphs show the results of fitting the time change in brightness variance to a mathematical formula; (a) is the unnormalized state, and (b) is the normalized state with the value at 0 min after the start of the test set at 1. The formula below each graph shows the results when the graph is fitted with that formula. a and b indicate values ​​specific to the food in question. k is the rate constant corresponding to the rate of change in the degree of digestion. R2 is the coefficient of determination; the closer it is to 1, the better the fitting results. [Figure 7](a) is an image taken 0 minutes after the start of the test, (b) is an image taken 0 minutes after the start of the test with black and white contrast enhanced by filtering, and (c) is a graph comparing the luminance variance with and without filtering. [Figure 8] (a) is an image taken approximately 180 minutes after the start of the test, (b) is an image taken approximately 180 minutes after the start of the test in which black and white contrast has been emphasized by filtering, and (c) is a graph comparing the luminance variance with and without filtering. [Figure 9] These are photographs taken at various times during a digestion test of mashed potato gel and carrots using a stomach model device. (a) is taken at 0 min, (b) is taken approximately 30 min, (c) is taken approximately 90 min, (d) is taken approximately 150 min, and (e) is taken approximately 180 min after the start. [Figure 10] The graphs show the time changes in texture features in grayscale images of color images during the digestion test of mashed potato gel and carrot, with (a) the change in variance of brightness, (b) the change in energy of brightness, and (c) the change in entropy of brightness. Each plot shows the results of a 40-second moving average of values ​​calculated at 1-second intervals, normalized with the value at 0 minutes from the start as 1. [Figure 11] (a) is a graph showing the Class_undigested and Class_digested images of mashed potato gel used to create a machine learning learning model, and (b) is a graph showing the number of times the images of Class_undigested and Class_digested mashed potato gel were trained and the probability of being Class_digested. The images in (a) show examples of 120 images taken at 1-second intervals from the start and end of the digestion test, looking forward and backward, respectively. The plot data in (b) show the average and standard deviation of five independent training results in a 5-fold cross-validation method. [Figure 12](a) is a photograph taken at each time point during a digestion test of mashed potato gel using a gastric model device, and (b) is a graph showing the time elapsed since the start of the digestion test of mashed potato gel based on the learning model and the probability of being classified as Class_digestion. Each plot of data in (b) shows the average value of the inference results at 1-second intervals for five independent learning models output by 5-fold cross-validation, which was then converted into a moving average value over 40 seconds. [Figure 13] (a) is an example of Class_undigested and Class_digested carrot images used to create a machine learning learning model, and (b) is a graph showing the number of times the images of Class_undigested and Class_digested carrots were learned and the probability of them being Class_digested. The images in (a) show examples of 120 images taken at 1-second intervals from the start and end of the digestion test, looking both forward and backward. [Figure 14] (a) is a photograph taken at each time point during a carrot digestion test using a stomach model device, and (b) is a graph showing the time elapsed since the start of the carrot digestion test based on the learning model and the probability of being classified as Class_digested. Each plot of data in (b) shows the average value of the inference results at 1-second intervals for five independent learning models output by the 5-fold cross-validation method, which was then converted into a moving average value over a 40-second period. BEST MODE FOR CARRYING OUT THE INVENTION

[0028] In the method of the present invention, as shown in Figure 1(a), first, the stomach body constituting the gastric model device is filled with artificial digestive fluid and food (mashed potato gel in this example) is placed into it. Next, the gastric model device is made to move in a manner similar to gastric peristalsis, and images (color video in this example) of the food digestion state over time are captured within a specific range. The capture range is selected to be one that allows the overall state of the solid-liquid mixture to be seen, including not only the solid portion (food) but also the liquid portion (background).

[0029] The digestive state is quantified by analyzing the texture of the captured image. Specifically, as shown in Figure 1(c), a brightness histogram is created and texture features are obtained from this histogram.

[0030] In this embodiment, as shown in Figures 2(a) to (c), a luminance histogram is created in the range to be analyzed, and the texture features obtained from this histogram are the average luminance, luminance variance, luminance contrast, luminance skewness, luminance kurtosis, luminance energy, and luminance entropy, and the degree of digestion is determined from the obtained texture features.

[0031] That is, a certain relationship is recognized between the degree of digestion and texture feature amount, and the degree of digestion can be known from the change over time in the texture feature amount. In the examples, the texture features mentioned are the luminance mean, luminance variance, luminance contrast, luminance skewness, luminance kurtosis, luminance energy, and luminance entropy. However, depending on the type of food, it is possible to determine the degree of digestion using only specific texture features without using all of these texture features.

[0032] The average luminance value (μ) is calculated using the following formula (1). The more the histogram is distributed towards higher luminance, the larger the value will be, and vice versa. In the following, i (eye) is the brightness value, p(i) is the frequency (probability) of brightness value i, N is the gradation, and entropy may have the base of log as the natural logarithm e.

number

[0033] Luminance variance (σ 2 ) is calculated using the following formula (2), and the value becomes larger as the histogram deviates from the mean value, and vice versa.

number

[0034] The luminance contrast is calculated using the following formula (3), and like the average value, the more the histogram is distributed towards higher luminance values, the larger the value will be, and vice versa.

number

[0035] Brightness skewness is calculated using the following formula (4) and indicates the symmetry of the histogram shape. The more symmetric it is about the mean value, the closer it is to 0. If there is a bias towards higher brightness values ​​relative to the mean value, the value will be negative, and if the opposite is true, the value will be positive.

number

[0036] The kurtosis of brightness is calculated using the following formula (5) and indicates the width of the tail of the histogram. If the tail is wider than the normal distribution, the value will be smaller than 3, and if the tail is wider than 3, the value will be larger.

number

[0037] The luminance energy is calculated using the following formula (6), and the more the histogram is concentrated at a specific luminance value, the larger the value will be, and vice versa.

number

[0038] The luminance entropy is calculated using the following formula (7), and the more diverse the luminance values ​​in the histogram, the larger the value will be, and vice versa.

number

[0039] Figure 3 shows photographs taken at various times during the digestion test of mashed potato gel using the stomach model apparatus; (a) is taken at 0 min, (b) is taken approximately 10 min, and (c) is taken approximately 180 min after the start. As can be seen from these photographs, the mashed potato gel was rapidly broken down into small pieces approximately 10 min after the start of the digestion test, and no significant changes were observed until approximately 180 min after the start. From this, it can be observed that the physical digestion of the mashed potato gel using the stomach model apparatus was completed approximately 10 min after the start.

[0040] Figure 4 shows the time change in texture features in the grayscale image of the color image during the digestion test. (a) is a graph showing the change in variance of brightness, (b) is a graph showing the change in energy of brightness, and (c) is a graph showing the change in entropy of brightness.

[0041] These graphs reveal significant changes in brightness variance, energy, and entropy approximately 10 minutes after the start of the digestion test. As the mashed potato gel breaks down during the digestion process, the food particles become uniformly distributed in the artificial digestive fluid. At this time, the dark, low-brightness artificial digestive fluid portions disappear from the image, and the bright, high-brightness portions resulting from the food particles take up the majority of the image. At this time, the histogram becomes concentrated at bright, high-brightness values, decreasing the variance. Energy, which indicates the degree of concentration at a specific brightness value, increases. The concentration of specific brightness values ​​reduces the diversity of brightness types, resulting in a decrease in entropy. Thus, the brightness variance, energy, and entropy changes reflect the degree of physical digestion of the mashed potato gel. Therefore, the variance, energy, and entropy changes can be considered effective texture features for determining the degree of physical digestion of the mashed potato gel.

[0042] Figure 5 is a graph showing the results of the same test as Figure 4, but with only the red channel extracted. This shows that similar texture analysis is possible even when the characteristic color channels are extracted.

[0043] Figure 6 shows graphs of the results of fitting the time change in luminance variance to a mathematical formula. (a) shows the graph without normalization, and (b) shows the graph with normalization. Normalization makes comparison and analysis easier.

[0044] In the present invention, texture features with a relatively high temporal resolution can be obtained according to the frame rate of the video (for a 60 fps video, the maximum resolution is 1 / 60 seconds). Therefore, by curve fitting the time change plot, the progress of gastric digestion can be quantified.

[0045] Figure 7 shows an example of preprocessing (filter processing) that emphasizes the feature values ​​of the image being analyzed. (a) and (b) show that the black-and-white contrast is different before and after processing, and that the variance of brightness, which is one of the texture feature values, is emphasized after processing.

[0046] Figure 8 shows an image taken approximately 180 minutes after the start of the digestion test (a) without preprocessing and (b) with preprocessing, and unlike the image taken 0 minutes after the start of the digestion test in Figure 7, no significant difference in brightness variance is observed, as shown in (c). In other words, filtering emphasizes the change in brightness variance between cases where physical digestion has not progressed and when it has progressed, making the results easier to see.

[0047] Figure 9 shows the difference in digestibility between mashed potato gel and carrot. Mashed potato gel is rapidly pulverized approximately 10 minutes after the start of the digestion test, with no significant change observed until approximately 180 minutes have passed. In contrast, carrots are barely pulverized until approximately 30 minutes after the start of the digestion test, then begin to be pulverized gradually around 90 minutes, and then rapidly pulverized between approximately 150 and 180 minutes. Furthermore, the degree of pulverization of carrots at approximately 180 minutes is lower than that of mashed potato gel.

[0048] As shown in Figure 10, the time changes in texture features show that for the mashed potato gel, the changes in brightness variance, energy, and entropy change significantly approximately 10 minutes after the start of the digestion test. In contrast, the carrots have higher overall variance and entropy levels than the mashed potato gel, and a higher energy level. The variance of the carrots shows almost no decrease until approximately 30 minutes after the start of the digestion test, then begins to gradually decrease around 90 minutes, and then decreases rapidly between approximately 150 and 180 minutes. Furthermore, the variance of the carrots at approximately 180 minutes is higher than that of the mashed potato gel. The change in entropy of the carrots over time shows a similar trend to the variance, albeit slightly. The energy content of carrots did not change significantly until approximately 150 minutes after the start of the digestion test, and then rose sharply between approximately 150 and 180 minutes. Furthermore, the energy content of carrots at approximately 180 minutes was lower than that of mashed potato gel. These trends correspond to the observation results in Figure 9.

[0049] Furthermore, the red, green, and blue channels were extracted and compared for the mashed potato gel and carrot. Because the mashed potato gel is white, it contains almost equal amounts of red, green, and blue, so there was almost no difference between the grayscale results and the red, green, and blue channel results. On the other hand, carrots are orange, with the most red, followed by green, and almost no blue, so if using color channels, it is preferable to use red or green. Another example of a case where a color channel would be used is when a food product loses a specific color (e.g., red) during digestion testing (is released into the digestive juice). By processing the food product in the channel of the color that loses color, it may be possible to quantify the color loss behavior.

[0050] Next, the digestion evaluation method using machine learning, which uses image brightness as a texture feature, is outlined as follows: Step 1: An image is imported into machine learning software obtained from the web. This image is made up of many pixels, each with a different brightness. Step 2: The captured image is divided into small regions and converted into feature data in which the local brightness characteristics of each small region are emphasized. Step 3: The feature data obtained in step 2 is converted into a one-dimensional vector. Step 4: The one-dimensional vector obtained in step 3 is converted to a scalar. Step 5: The scalar obtained in step 4 is subjected to a sigmoid function to transform it so that its minimum value is 0 and its maximum value is 1. Step 6: Finally, a number between 0 and 1 indicating the degree of digestion for the input image is output (displayed).

[0051] Next, an example of a method for evaluating the digestibility (degree of digestion) of mashed potato gel based on machine learning will be described with reference to Figures 11 and 12. In this example, the start of the digestion test using the stomach model device shown in Figure 3 corresponds to Class_undigested, and the end of the test corresponds to Class_digested.

[0052] Using available machine learning software (e.g., ResNet, EfficientNet), create a learning model in which the value representing Class_Undigested is 0 and the value representing Class_Digested is 1.

[0053] As data for creating a learning model, in the case of mashed potato gel, 120 images were prepared at the start of the digestion test and 120 images at the end of the test, as shown in Figure 11(a). Of the 120 images prepared, 96 (4 / 5) were used as training data and 24 (1 / 5) were used as validation data, as the 5-fold cross-validation method was later applied. The learning data was subjected to the processing of steps 1 to 6. During this process, the coefficients of the arithmetic formulas used in steps 2 and 4 were adjusted so that the numerical value of step 6 for the image at the start of the digestion test would be close to 0, and so that the numerical value of step 6 for the image at the end of the test would be close to 1. In this example, the above verification data was used to calculate scores for the undigested and digested images in step 6. Steps 1 to 6 were repeated until the score for the undigested image was 0.1 or less and the score for the digested image was 0.9 or more.

[0054] Figure 11(b) is a graph showing the relationship between the number of learning sessions and the probability of Class_ digestion. From this graph, we can see that as the number of learning sessions increases, the score for undigested images approaches 0, and the score for digested images approaches 1.

[0055] Figure 12(a) shows photographs taken at each time point during the mashed potato gel digestion test, and Figure 12(b) shows the time elapsed since the start of the mashed potato gel digestion test and the probability of being Class_digested based on the learning model. As is clear from Figure 12(b), in the case of mashed potato gel, digestion progressed rapidly (with a probability of Class_ digestion of about 80%) from the start of the digestion test up to 30 minutes, after which digestion progressed more slowly. This trend corresponds to the results without machine learning.

[0056] 13 and 14 show the results of a learning model created for carrots, similar to the mashed potato gel, and a digestibility evaluation. Figure 13(a) shows the images (120 images) at the start of the digestion test to create the learning model and the images (120 images) at the end of the test. Figure 13(b) is a graph showing the relationship between the number of training sessions and the probability of Class_digestion. As with the mashed potato gel example, as the number of training sessions increases, the score for undigested images approaches 0, and the score for digested images approaches 1.

[0057] As can be seen from Figure 14(a), in the case of carrots, the peristaltic movement of the stomach model device only gradually breaks down the carrots compared to the mashed potato gel, and the probability of Class_ digestion does not increase as rapidly as in the case of the mashed potato gel, but rather increases in proportion to the elapsed time. Furthermore, as shown in Figure 14(b), the probability of being classified as Class_digestible does not increase rapidly over the first 30 minutes of the digestion test, as was the case with mashed potato gel, but rather increases gradually over the time elapsed since the start of the digestion test. This trend corresponds to the results without machine learning.

[0058] As the degree of digestion changes over time for each food item, it is necessary to create a machine learning model for each food item. [Industrial Applicability]

[0059] Mashed potato gel and carrots were used in the digestion test examples, but the present invention can be applied to other foods. However, since the physical characteristics of foods differ depending on the food, the rate at which physical digestion progresses varies for each food. However, by capturing the temporal changes in the texture features described above, the progress of physical digestion for each food can be quantitatively determined.

Claims

1. A digestion evaluation method that performs texture analysis on a color image of food obtained from a stomach model device, characterized in that the texture analysis quantifies the degree of digestion based on temporal changes in at least one of the image's luminance mean, luminance variance, luminance contrast, luminance skewness, luminance kurtosis, luminance energy, and luminance entropy.

2. 2. The digestion evaluation method according to claim 1, wherein the texture analysis is performed using color channels corresponding to the type of food.

3. 2. The digestion evaluation method according to claim 1, wherein the image is filtered to enhance luminance levels when the texture analysis is performed.

4. This is a digestion evaluation method that performs texture analysis on a color image of food obtained from a stomach model device, capturing the brightness information of the color image of the food as a texture feature, having a computer learn the texture feature values ​​at the start and end of a digestion test as undigested and digested states, respectively, to create a machine learning learning model, and using this learning model to quantify the degree of digestion of the image to be measured.

Images