Method and system for characterizing infant feces pattern

Abstract

The present invention provides a method of analyzing the consistency of faeces, comprising the steps of providing an infant's faeces, capturing an image of the faeces with a portable device comprising a camera, providing the captured image to an input layer of a pre-trained convolutional neural network (CNN), processing the captured image using the CNN, thereby obtaining a classification vector from a final layer of the CNN, and obtaining information about predicted scores from the classification vector, where at least a final layer of the CNN is customized such that each element of the classification vector corresponds to one score of a faeces analysis scale, and storing the information about the predicted scores.

Description

[0001] This application is a divisional application of the invention patent application with the application date of May 28, 2019, application number 201980040963.5, and name “Methods and Systems for Characterizing Infant and Young Child Stool Patterns”. Technical Field

[0002] The present invention relates to a method and system for analyzing the consistency of a stool sample. Background Art

[0003] The health care of an infant is the most important task for parents. Advances in technology allow for tasks that assist parents and caregivers by tracking the development of an infant and quickly detecting certain anomalies. When it comes to infant nutrition, it is important to detect if there are any anomalies with the infant's digestive system, or if the infant's body is absorbing all the nutrients it needs.

[0004] In order to assess the capacity of the digestive system, it is known that analyzing stool patterns can provide good insights. The scale obtained by comparing stool with a set of stool analysis scale scores helps to classify the type of stool and retrieve conclusions therefrom. The embodiments of this scale are the Bristol Stool Morphology Scale (BSS) and the Amsterdam Stool Scale. The BSS is composed of seven stool images of different consistencies, allowing to objectively assess the consistency of adult stool (scale 1 represents hard agglomerated stool until 7 represents watery stool). The BSS can also be used to characterize the stool of infants and young children.

[0005] When parents visit health care professionals (HCPs) because they are concerned about their infant's health, HCPs often ask questions about the consistency of their infant's stool that are difficult for most parents to answer. When parents are asked to record their infant's stool consistency, they have difficulty identifying stool consistency and stool analysis scale scores that correlate with their child's stool.

[0006] It is desirable to have a system in which parents and caregivers can keep track of their infant's stool pattern (i.e., stool consistency, frequency, and color) in real time. It is also desirable to track the stool pattern in an objective and consistent manner, regardless of which caregiver (parent, grandparent, babysitter, or daycare caregiver) is changing a diaper, or helping a child use a potty or commode chair. It is also desirable to have a system that provides an indication based on the observed stool pattern, either indicating that everything is normal, which will reassure parents and caregivers, or indicating that the infant's stool pattern is not consistent with expectations, and a consultation with an HCP is recommended.

[0007] Nowadays, as people have the habit of constantly carrying their smartphones, tablets or other portable devices, there are efforts to provide mobile applications (apps) performed through these portable devices, which can make daily tasks easier for users. In addition, the method of taking pictures of babies in order to observe their development is a common practice for parents, with the help of which they can roughly understand the growth of their babies.

[0008] The above method can also be used to keep track of stool patterns. Known programs or applications allow for importing or capturing images of stool and manually selecting the score of the stool analysis scale that is more appropriate for the stool on the image in order to keep a record of the digestive system's capabilities over time. In addition, known programs or applications allow for the automatic detection of stool color using color recognition technology.

[0009] While these applications can help parents and caregivers analyze some characteristics of their child's stool, sometimes manual classification is difficult and manual record keeping is tedious, so a method is needed that can more accurately analyze the characteristics of a baby's stool in a manner that is simple for parents and caregivers and also provides accurate and quick classification. Summary of the invention

[0010] The present invention provides a method for analyzing stool consistency, comprising the following steps: providing infant stool, capturing an image of the stool using a portable device including a camera, providing the captured image to an input layer of a pre-trained convolutional neural network (CNN), processing the captured image using the CNN to obtain a classification vector from a final layer of the CNN, and obtaining information about a predicted score from the classification vector, wherein at least the final layer of the CNN is customized so that each element of the classification vector corresponds to a corresponding score of a stool analysis scale, and storing information about the predicted score.

[0011] Thus, the present invention provides a method for characterizing infant feces that is easy to implement and provides fast and accurate classification results. Using CNN to classify feces provides improved accuracy for the classification results, and using the camera of a portable device to capture the image allows the task of the parent or caregiver to be simplified, as the image of the feces can be simply captured and the results automatically received without further manipulation of the feces. In addition, there is no inter-observer variability, as the results will be provided in an objective manner regardless of who captures the image.

[0012] In one embodiment according to the invention, the excrement is disposed in an open diaper. Since the image can be captured directly from the diaper in which the excrement is disposed, this allows simplifying the task of parents and caregivers, and when changing the diaper of the baby, the image can be easily captured before handling the diaper.

[0013] In one embodiment according to the invention, the stool is placed in a potty, a bed commode, a toilet chair or a toilet with a platform. Thus, the invention facilitates the task of parents and caregivers by eliminating the need for them to manipulate the stool, since the image can be captured directly at the location where the baby places the stool (suitable for different possible situations).

[0014] In one embodiment according to the present invention, the method further comprises, before capturing the image, displaying guidance information by the portable device so as to meet the predetermined condition. This can help the user of the portable device (e.g., a parent, other family member or caregiver) to capture an image with optimal features to be used in the classification step.

[0015] In one embodiment according to the present invention, capturing the image includes automatically capturing the image by the portable device when a predetermined condition is met.

[0016] In one embodiment according to the present invention, the predetermined condition includes a condition that the feces is placed on a surface with a regular background. The regular background can be a background with a specific pattern, a uniform background, etc. In one embodiment according to the present invention, the predetermined condition includes a condition that the feces is the only object appearing in the image.

[0017] If the feces is set in an opened diaper, in an embodiment according to the present invention, the predetermined condition includes the condition that the diaper is placed on a surface with a regular background, and in one embodiment according to the present invention, the predetermined condition includes the condition that the diaper is the only object appearing in the image.

[0018] In one embodiment according to the invention, the information about the predicted score includes at least one of the following information: the predicted score, the captured image of the stool, and the date and / or time when the image was captured. The information may also include other types of information useful to parents and caregivers or to the application, such as whether the color of the stool is considered normal.

[0019] In one embodiment according to the invention, processing of the captured image using the CNN is performed by a portable device. In one embodiment according to the invention, processing of the captured image using the CNN is performed by a server in communication with the portable device. In this second case, the portable device can send the captured image to a server in communication with the portable device via a network (e.g., the Internet) and receive a classification vector or a predicted score from the server. This may be advantageous in cases where the portable device does not have sufficient computing power to perform all the operations required in the CNN. However, the first case may be advantageous when the portable device is not connected to a network because it allows offline prediction of scores.

[0020] In one embodiment according to the invention, information about the predicted scores is stored in at least one of the portable device or a server in communication with the portable device. The server can be used to store all information about the predicted scores and can also be used to store the captured images, thereby freeing up memory from the portable device and allowing, for example, the HCP to access the information if necessary. The portable device can also locally store a copy of the captured images and information (scores) derived from the captured images.

[0021] In one embodiment according to the invention, the image is a color image. This allows the image to contain more information (third level information), which can be used for CNN to improve its performance. In one embodiment according to the invention, the method also includes automatically detecting the color of feces from the captured image and analyzing it to provide information. This can allow the method to also provide information about whether the color is normal, which, together with the consistency, can allow better determination of possible abnormalities.

[0022] In one embodiment according to the invention, the method further comprises analyzing the frequency of stool generation based on storage time information associated with the prediction scores of the plurality of captured images. The frequency can be used to provide additional information by determining the frequency of occurrence based on the number of images captured per day and the time at which they are captured. In addition, the combination of consistency and frequency can be used to provide information such as diarrhea (watery stool three or more times a day) or constipation (hard stool twice a week or less).

[0023] The present invention also provides a system for analyzing stool consistency, the system comprising: a portable device, the portable device comprising: a camera, the camera being configured to capture an image of an infant's stool; a controller, the controller being configured to provide the captured image to an input layer of a pre-trained convolutional neural network (CNN), obtain a classification vector from a final layer of the CNN, wherein at least the final layer of the CNN is customized so that each element of the classification vector corresponds to a corresponding score of a stool analysis scale, and obtain information about a predicted score from the classification vector; and a memory for storing information about the predicted score.

[0024] In one embodiment according to the present invention, the feces is arranged in an opened diaper.In one embodiment according to the present invention, the feces is arranged in a bedpan, a bed commode, a toilet chair, or a toilet with a platform.

[0025] In one embodiment according to the present invention, the information about the predicted score includes at least one of the following information: the predicted score, the captured image of stool, and the date and / or time when the image was captured.

[0026] In one embodiment according to the present invention, the system further comprises a server for communicating with the portable device, wherein the server is configured to store information about the predicted score.

[0027] In one embodiment according to the present invention, processing the captured image using the CNN to obtain a classification vector from a final layer of the CNN is performed by a portable device.

[0028] In one embodiment according to the present invention, the server is further configured to process the captured image using a CNN, thereby obtaining a classification vector from the final layer of the CNN, and the portable device is further configured to transmit the captured image to the server and receive the predicted score from the server. Thus, in one embodiment, the server can perform a determination of the most likely score among the possible scores of the stool analysis scale based on the classification vector and send the score to the portable device.

[0029] In one embodiment according to the present invention, the image is a color image.

[0030] The present invention also provides a computer program product for analyzing the consistency of feces, comprising a computer-readable medium, wherein the computer-readable medium comprises codes executable by at least one processor, thereby performing the method according to the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present invention will be discussed in more detail below with reference to the accompanying drawings, in which:

[0032] Figure 1 An overview of one embodiment of the present invention is described.

[0033] Figure 2A and Figure 2B Different disposals of diapers containing faeces according to the invention are shown.

[0034] Figure 3 A flow chart illustrating a method according to the invention is shown.

[0035] Figure 4 A diagram of a training image according to the present invention is shown.

[0036] Figure 5 A portable device according to an embodiment of the invention is schematically shown.

[0037] Figure 6 A portable device and a server according to an embodiment of the present invention are schematically shown. DETAILED DESCRIPTION

[0038] Figure 1 An overview of one embodiment of the present invention is depicted. Although embodiments of the present invention include stool placed in a diaper, bedpan, bed commode, commode chair, toilet with platform or other similar locations, Figure 1 In an embodiment, an example of feces being placed in a diaper is described. A diaper 20 containing baby feces 30 is placed on a surface in an open position, and an image of the opened diaper 20 with feces 30 is captured by a user using a portable device 10. The captured image can be used by an application running on the portable device, and the application can input the captured image into a model including a CNN, which will perform a series of operations to obtain a classification vector having probabilities of possible scores in the feces analysis scale, thereby obtaining a predicted score (i.e., a score with the highest probability) based on the classification vector. In order to obtain the best results from the classification process, it is expected that the captured image has good features, and therefore, certain predetermined conditions need to be met. As an example of a condition that needs to be met, the feces 30 should be as recent as possible so that its characteristics (color, consistency) have not changed due to, for example, partial absorption of the feces by the diaper. Therefore, the image should be captured shortly after the feces are discharged, and in Figure 1In an embodiment, the image should be captured shortly after the diaper 20 is filled with feces, and a suitable period for capturing the image is within ten minutes from the time the diaper is filled. The color of the feces in the captured image can also be used by CNN or a different algorithm to provide additional information. The application of the portable device can automatically detect the color of the feces from the captured image and analyze it to provide information. This also allows information to be provided regarding whether the color is normal, which together with the consistency can allow a better determination of possible abnormalities.

[0039] Another example of a condition that should be met is to capture the image with enough light to clearly distinguish features in the image, but not too much light as this may change the true color and appearance of the image features. One example of how to achieve this is to use natural light (daylight) or light from ceiling lights. However, using a camera's flash is less preferred as this may change the appearance of the image.

[0040] Another example of a condition is that the background of the image should preferably be regular, such as following a pattern or being uniform. If the stool is placed in a bedpan, bed commode, commode chair, or toilet with a platform, then it is desirable that the background be uniform. If the stool is placed in a diaper (such as Figure 1 ), then preferably the diaper 20 is in the foreground of the image. A uniform table surface is an example of a suitable background.

[0041] Another embodiment of the condition that should be met is that no other objects should appear in the image. If feces is placed in a bedpan, a bed commode, a toilet chair, or a toilet with a platform, it is expected that only feces and background will appear in the image, without other objects or parts of the body. If feces is placed in a diaper, it is expected that no other objects will appear except the diaper 20 containing feces 30. Parts of the body of infants or other objects should not appear in the image. However, if the captured image has undesirable objects around the feces, the captured image can be cut to eliminate undesirable objects before the image is input to CNN. Other pre-processing steps can be performed, such as modifying the resolution of the captured image, changing the format of the image, or other steps of removing noise in the image.

[0042] It should be noted that embodiments of the present invention may use only one of the predetermined conditions, or any combination thereof. It should also be noted that other predetermined conditions may be used as long as they help the user obtain an image suitable for input to a CNN. The predetermined conditions to be used may be determined and modified in the settings of the application, or may be predefined, and the controller of the portable device may control the predetermined conditions to be used, and how these predetermined conditions are determined.

[0043] Other examples of conditions that should be met can be found in Figure 2A and Figure 2B It was observed that Figure 2A and Figure 2B Different disposals of diapers containing feces according to the present invention are shown. Again, an embodiment of feces in a diaper is shown, but the skilled person will appreciate that these features can be similarly applied to other embodiments where feces are disposed in a bedpan, a bed commode, a commode chair, a commode with a platform, etc. Figure 2A In FIG. 1 , an opened diaper 20 including feces 30 can be observed with a uniform background 40, which in this case corresponds to the top of a table of uniform surface. Figure 2A In the image, there are no objects other than diapers. Figure 2A Such images may be considered suitable for the method of the present invention.

[0044] According to one embodiment, before capturing an image, for example, before pointing the camera at the feces, or when the user points the camera at the feces, the portable device 10 can issue instructions to the user to remind him / her of some conditions that should be met (sufficient light, etc.). These conditions can be some or all of the above-mentioned predetermined conditions, or different conditions. The portable device can additionally guide the user to move the portable device 10 closer to or farther away from the feces. Figure 2A In one embodiment, the portable device 10 can automatically capture an image when the portable device 10 detects that the portable device 10 is located at a suitable distance from the feces or diaper 20, when the light conditions are suitable, or when any other desired conditions are met. One or more conditions that facilitate the portable device to decide to automatically capture an image can be pre-defined.

[0045] exist Figure 2B In the embodiment, since the diaper 20 is not fully opened, the feces 30 cannot be adequately identified, and therefore the diaper 20 is not in a position suitable for capturing an image. According to one embodiment of the present invention, by observing the guidance displayed in the portable device, the user can be made aware that this is an inappropriate position and can open the diaper. According to another embodiment, the application will provide guidance before pointing the camera at the feces, and the user will be aware that, for example, the diaper needs to be opened so that the image is not captured. Figure 2B An image like the image represented in .

[0046] Figure 3A flow chart illustrating a method according to the present invention is shown. In step 301, feces 30 are provided. As described above, feces 30 can be provided in a diaper, which is then opened to capture an image, or feces 30 can also be provided in a bedpan, a bed commode, a commode chair, or a commode with a platform. In step 302, an image of feces 30 is captured using a camera included in the portable device 10. The capturing step can be performed by an application running in the portable device 10, which can be initialized by the user.

[0047] In step 303, the captured image is provided to the input layer of a convolutional neural network (CNN). CNN is a neural network suitable for classifying images, and unlike conventional neural networks, the layers of CNN have neurons arranged in three dimensions (width, height, and depth). For the purposes of this disclosure, CNN is understood to be a neural network with multiple layers, such as a feedforward neural network, which includes a layer that converts an input 3D quantity into an output 3D quantity. In one embodiment, CNN includes an input layer and an output layer, with multiple hidden layers between the input layer and the output layer. Each hidden layer can be one of a convolutional layer, a pooling layer, a fully connected layer, and a normalization layer. For example, such a neural network can be implemented using the TensorFlow library (Abadi et al., "TensorFlow: A System for Large-Scale Machine Learning", the 12th USENIX Symposium on the Design and Implementation of Operating Systems, 2016).

[0048] Throughout the various layers of the CNN, the complete input image (color or grayscale) is simplified into a single vector, or classification vector, for classification scores. In the present invention, step 304 includes processing the captured image to obtain a classification vector and obtaining a predicted score based on the stool analysis scale. The image captured with the portable device is the input image of the CNN, which can be a red, green, blue image or a grayscale image. The single vector output is a classification vector including probability values ​​for each possible score. The predicted score can be set to the score with the highest probability in the score of the classification vector.

[0049] Several stool analysis scales can be used, for example the Bristol Stool Morphology Scale (BSS), the Amsterdam Stool Scale, or any other suitable scale, such as the Brussels Infant and Toddler Stool Scale (BITSS) currently under development (VandenPla et al., Development of the Brussels Infant and Toddler Stool Scale (“BITSS”): Study Protocol, BMJ Open 2017;7:e014620).

[0050] For example, if BSS is used, the vector is a 1x1x7 vector containing the probability of each of the seven possible scores that make up the scale.

[0051] Step 304 may be performed by a controller of the portable device 10, or may be performed by a server communicating with the portable device 10. In the second case, the controller of the portable device will instruct the transceiver of the portable device 10 to transmit the captured image to the server, and the server will perform the processing of the captured image, that is, the calculation of the CNN, thereby providing a classification vector and / or a predicted score. This is applicable when the portable device lacks sufficient computing power and graphics capabilities to perform the operations required by the CNN. After obtaining the classification vector by the server or the portable device 10, step 305 consists of storing information about the classification vector or about the score. The information may be stored in the portable device 10 or in the server, or in both. If the information about the classification vector is stored, the portable device can then make a decision on the most likely score by selecting the score with the highest probability value in the scores of the classification vector. If the score is stored, the step of making the decision can be omitted. The score is then displayed by the portable device 10 executing the application. In another embodiment, the server can also perform the selection of the score based on the probability values ​​of all scores and send the score to the portable device. Information about the predicted score includes at least one of: the score itself, the captured image, information about the date and / or time the image was captured, a classification vector obtained by the CNN, or some additional information that may be useful to parents and caregivers and the application.

[0052] Figure 4A diagram of training images according to the present invention is shown. When an application for providing feces information is executed in a portable device 10, the captured image is input into a model including a CNN, and as a result, the CNN will provide a classification vector, or a predicted score from the classification vector, the classification vector including a probability value of a possible score within the feces analysis scale. In order to provide the classification vector, the CNN needs to be pre-trained in advance, that is, it is necessary to provide the CNN with enough information to be able to classify the input image. This can be performed by a collection of labeled images (labeled with scores of the feces analysis scale) so that from the collection of labeled images, the CNN can learn to automatically predict specific labels. Transfer learning can also be used to answer specific image recognition tasks (such as predicting BSS or BITSS scores from pictures of feces). This method allows the establishment of a model that can answer specific questions without collecting thousands of pictures and training the model from scratch. In this method, a pre-trained model can be used and only the last few layers can be customized to predict scores. The CNN to be used can be based on a known CNN (e.g., based on the Tensorflow library), but at least the last layer is customized so that the output of the CNN provides a classification vector containing each score of the stool analysis scale. The training of at least the last layer of the CNN is performed as follows. In step 401, an initial set of images is provided. An embodiment of a suitable set size is at least 200 images per score, at least 1000 images in total (including images of all scores of the scale), more preferably at least 1500 images, and more preferably at least 1600 images. These images are manually labeled with a score of the stool analysis scale (such as BBS) used. In the initial set of images, a subset is selected as a test image in step 402, and a subset is selected as a training image in step 403. Through the subset of training images, a training model is implemented in step 404, and in the training model, the values ​​of certain elements in the operations performed are updated by comparing the results of the operations with the labels that the images originally had. In this way, CNN "learns" how to better predict the correct score of the image. After this, a subset of test images is used to evaluate the model in step 405. Once this iteration is completed, step 406 may optionally be performed, where the size of the training set may be increased (and therefore the size of the test set may be reduced), and the iterative process may be repeated starting from step 401. This fine-tuning step 406 may increase the accuracy of the model. Once the CNN is trained, the values ​​of its elements used in different operations are known, and with these values, the CNN is able to predict the classification of an input image (such as an image of feces in a diaper captured by a portable device). Subsequently, as described above, the trained model containing the CNN may be run in a portable device or server to provide a score for the captured image.According to an embodiment of the present invention, in addition to the fine-tuning step 406, the application may allow for the collection of additional images for retraining the model if the initial data set is insufficient to achieve acceptable accuracy.

[0053] During the training step, and in order to improve the accuracy and reliability of the classification, the following steps may be performed: wherein for the manual labeling of the images, an assessment is given by at least one of the parents (preferably both) or a caregiver (e.g., the mother), wherein preferably these parents or caregivers have no relationship with the infant from whom the initial set of images was obtained. In a preferred embodiment, when both parents or caregivers (or a combination) are used for the assessment, any disagreement in their assessment is resolved by the assessment conclusion of a healthcare professional in order to obtain a final score, i.e., a label.

[0054] In an example according to an embodiment of the present invention, a total of 2731 images are used as an initial set of images. Among these images, a subset of 2478 images is selected as training images, and a subset of 209 images is selected as test images. Table 1 below shows an example of performing evaluation using 209 test images.

[0055]

[0056] Table 1

[0057] As shown in Table 1, out of the 209 test images, 16 images were identified as belonging to score 1 by both humans and machine learning classification systems. Similarly, as shown in the cells in the diagonal of Table 1, 10 images were identified as belonging to score 2 by both humans and machine learning classification systems. This represents a level of agreement of 63.6%. This level of agreement is even higher if an error of ±1 point is considered, as shown in the cells adjacent to the diagonal elements in Table 1, from which it can be seen that the agreement reaches 93.8%.

[0058] According to an embodiment of the present invention, the classification of stool consistency provided by an application running in a portable device or on a server follows the seven possible scores of the BSS.

[0059] In one embodiment according to the present invention, the operating system of the portable device is iOS and / or Android. The application for providing stool classification can be run in the portable device 10 as follows:

[0060] First, there may be a registration process. When the application is opened, the user is provided with the option to log in or register. If register is clicked, a new screen may be displayed to request a code (such as a one-time password (OTP) code). Once entered and verified as valid by the system, a new screen may be displayed asking for login credentials, such as an email and password. After that, the baby's information (name, date of birth, gender) may be imported in the next screen.

[0061] Secondly, once registration is complete, or once the user logs in, a "home" screen with different types of information and options can be displayed, including information about the "feces module". In the feces module, the user can create a new entry, thereby being allowed to capture images as explained above, or enter a stored image, which can be previously captured or received by the portable device. According to one embodiment of the present invention, the portable device can automatically capture images when certain conditions are met. According to another embodiment, the application can display a certain guide so that the user can take photos with good enough features. The guide can be information such as "In order to obtain the best quality image, please be located in a room with enough light and try to capture only feces (and diapers)". However, the technician understands that this is only an embodiment and other guide information can be displayed. Subsequently, the application will display the scores of the feces analysis scale (such as, for example, based on the water sample, soft, formed and hard corresponding to the BSS) using the above-disclosed method, and the information related to the score can be stored in the portable device. Similarly, the captured image can also be stored in the portable device. Where multiple entries are stored, a progression graph may be created to allow tracking of changes in stool consistency and also facilitate determination of frequency of stool provision, which may provide additional information relating to, for example, diarrhoea or constipation.

[0062] Entries can be modified and deleted. These entries can also be shared with a health database (server) shared with the infant's health care provider (HCP) so that if there is an indication for intensive care, the portable device can alert the user to contact the HCP, who can then quickly access the relevant information, allowing for a faster decision to be made on the next step.

[0063] The HCP can compare the predicted scores with the actual images and, if necessary, correct the scores. In one embodiment, such corrected samples are automatically entered into another set of training data so that the model can be retrained to improve accuracy.

[0064] All images captured using the application may be stored on the portable device and / or on a remote server that may communicate with the mobile device, which would allow the HCP to access the data and evaluate their patient's data if access is granted.

[0065] Figure 5 A portable device 10 according to one embodiment of the present invention is schematically shown. The portable device 10 has a display unit 501, which may be a touch screen suitable for displaying information and processing user input. The device 10 also has a camera 502 for recording images and video clips, a processor 503 for processing recorded images, a memory 504 for storing images, program data, CNNs, etc., and a communication unit 505 for communicating with other devices via a wired connection or a wireless connection. In one embodiment, the processor 503 is programmed to process the recorded images using a CNN and generally implement the processes described in this application.

[0066] Figure 6 A portable device 10 and a server 100 according to an embodiment of the present invention are schematically shown. The portable device 10 and the server 100 can communicate via a wired connection or a wireless connection. In one embodiment, the portable device 10 sends a recorded image to the server. The server has a processor, a memory and a communication unit. The server 100 can be programmed to process the received image using a CNN and send the result back to the portable device 10. In addition, the server can store the obtained results and / or store the received image and / or store any intermediate calculation results. The server can also be arranged to implement a reference Figure 4 The training method described.

[0067] In the foregoing description of the drawings, the invention has been described with reference to specific embodiments thereof. However, it will be apparent that various modifications and changes may be made to the invention without departing from the scope of the invention as summarized in the appended claims.

[0068] In particular, specific features of various aspects of the invention may be combined. One aspect of the invention may be further advantageously enhanced by adding features described in relation to another aspect of the invention.

[0069] It should be understood that the present invention is limited only by the appended claims and their technical equivalents. In this document and its claims, the verb "comprise" and its variations are used in their non-restrictive sense, meaning to include the items after the word, without excluding items not specifically mentioned. In addition, the elements mentioned by the indefinite article "a" or "an" do not exclude the possibility of more than one of the elements, unless the context clearly requires only one element. Therefore, the indefinite article "a" or "an" usually means "at least one".

Claims

1. A computer-implemented method for analyzing stool consistency, comprising the steps of: - Provide baby's feces, - Capturing images of the feces with a portable device that includes a camera, -Provide the captured image to the input layer of a pre-trained convolutional neural network (CNN); - processing the captured image using a CNN by the portable device, thereby obtaining a classification vector from a final layer of the CNN, and obtaining information about a predicted score from the classification vector; wherein the CNN has been obtained by transfer learning by customizing at least the final layer of the CNN such that each element of the classification vector corresponds to a respective score on a stool consistency analysis scale, and - store information about the predicted scores, - wherein the image is a color image, and wherein the method further comprises automatically detecting, by the portable device, the color of the stool, and analyzing the color to provide additional information.

2. The method of claim 1, wherein the feces is disposed in one of an open diaper, a bedpan, a bed commode, a commode chair, or a toilet with a platform.

3. The method according to any one of the preceding claims, further comprising displaying guidance information by the portable device to satisfy a predetermined condition before capturing an image. 4 . The method according to claim 1 , wherein capturing an image comprises automatically capturing an image by the portable device when a predetermined condition is met.

5. A method according to claim 3 or 4 when dependent on claim 2, wherein the predetermined condition includes at least one of the following conditions: a condition that feces or diapers are placed on a surface with a regular background, or a condition that feces or diapers are the only objects appearing in the image.

6. The method according to any of the preceding claims, wherein the information about the predicted score comprises at least one of the following information: the predicted score, the captured image of the stool, and the date and / or time when the image was captured.

7. A method according to any preceding claim, wherein information about the predicted score is stored in at least one of the portable device or a server in communication with the portable device.

8. The method according to claim 1, further comprising analyzing a frequency of stool generation based on storage time information associated with the prediction scores of the plurality of captured images.

9. A system for analyzing stool consistency, comprising: - A portable device, comprising: a camera configured to capture images of feces of the infant, A controller, the controller being configured to: The captured image is fed into the input layer of a pre-trained convolutional neural network (CNN). processing the captured image using a CNN to obtain a classification vector from a final layer of the CNN; wherein the CNN has been obtained by customizing at least the final layer of the CNN through transfer learning so that each element of the classification vector corresponds to a respective score of a stool consistency analysis scale, and obtaining information about the predicted scores from the classification vector, and a memory for storing information about the predicted scores, Wherein the image is a color image, and wherein the controller is configured to automatically detect the color of the stool and analyze the color to provide additional information.

10. A system according to claim 9, configured to operate according to any one of claims 2 to 8.

11. The system of any one of claims 9 or 10, further comprising a server for communicating with the portable device, wherein the server is configured to store information about the predicted score.

12. A computer program product for analyzing stool consistency, comprising: A computer-readable medium comprising codes executable by at least one processor to perform the method according to any one of claims 1 to 8.

Images