Program and information processing device

Abstract

The present invention accurately evaluates the condition of a bedsore. A computer is caused to execute: acquiring an affected part image in which an affected part is captured; inputting the affected part image to a first learning model trained using first teacher data in which an affected part image for learning different from the affected part image is associated with the presence or absence of skin damage in the affected part image for learning, to acquire the presence or absence of skin damage in the affected part image; inputting an input image based on the affected part image to a second learning model trained using second teacher data in which a damaged affected part image for learning, which is an image in the affected part image for learning that includes an affected part in which skin damage is present, is associated with classification information for calculating a level classification based on the level of bedsore in the damaged affected part image for learning; and acquiring classification information for calculating the level classification of bedsore in the affected part image from the second learning model.

Description

Program and Information Processing Apparatus

[0001] The present invention relates to a program and an information processing apparatus.

[0002] In a medical or nursing care setting, a patient may be bedridden, for example, and pressure ulcers (bedsores) may develop at sites compressed by body weight. Depending on the state of the pressure ulcer, there are various care methods for the pressure ulcer, and it is necessary to evaluate the state of the pressure ulcer in order to select an appropriate care method. When evaluating the state of a pressure ulcer, Patent Document 1 describes an imaging device that enables appropriate imaging of an affected area when photographing a pressure ulcer.

[0003] Japanese Patent Application Laid-Open No. 2020-156082

[0004] The evaluation of the state of pressure ulcers, including the example of Patent Document 1, is performed based on the DESIGN-R (registered trademark) 2020 classification, which is a pressure ulcer state determination scale developed by the Academic Education Committee of the Japanese Pressure Ulcer Society. An example of the evaluation items according to the DESIGN-R 2020 classification is shown in FIG. 8. In the DESIGN-R 2020 classification, there are many evaluation items, and there are not a few differences in evaluation even among skilled medical workers. Furthermore, it is even more difficult for caregivers of homebound patients and medical staff in疗养 hospitals to perform an appropriate evaluation of the state of pressure ulcers.

[0005] Therefore, an object of the present invention is to provide a program and an information processing apparatus capable of accurately evaluating the state of pressure ulcers.

[0006] A program according to one aspect of this disclosure causes a computer to perform the following actions: acquire an image of the affected area taken; input the image of the affected area into a first learning model trained using first training data linked to a different training image of the affected area and the presence or absence of skin damage in the training image of the affected area, and acquire the presence or absence of skin damage in the image of the affected area; and input an input image based on the image of the affected area into a second learning model trained using second training data linked to a training image of the affected area that includes an image of the affected area with skin damage, and classification information for calculating a pressure ulcer level classification based on the level of pressure ulcer in the training image of the damaged area, and acquire classification information for calculating a pressure ulcer level classification in the image of the affected area from the second learning model.

[0007] This allows the first learning model to remove images without skin damage, which can reduce accuracy when classifying pressure ulcer levels based on images of the affected area. By inputting the input images, which the first learning model has determined to have skin damage, into the second learning model, classification information for calculating pressure ulcer levels can be obtained with high accuracy. Since the pressure ulcer level can be evaluated based on the accurately calculated classification information, the condition of the pressure ulcer can be evaluated with high accuracy.

[0008] Furthermore, in the program according to the above embodiment, the pressure ulcer level classification is a depth level classification, which is a level classification based on the depth of the pressure ulcer, and the classification information may be a probability indicating whether the depth of the pressure ulcer falls into one of the depth level classifications.

[0009] This makes it possible to accurately calculate the level classification of pressure ulcers based on their depth.

[0010] Furthermore, in the above embodiment, the program may also cause the computer to input the injured area image into a third learning model, which has been trained using third training data linked to the size information of the pressure ulcer in the injured area image, and to obtain the size information of the pressure ulcer from the third learning model, and to obtain classification information from the second learning model, which may also include obtaining classification information from the second learning model using input images extracted from the injured area image based on the size information as input.

[0011] This allows the system to use images extracted based on the size of the pressure ulcer as input, thereby suppressing the decrease in accuracy caused by inputting parts other than the pressure ulcer, and enabling accurate calculation of the level classification in relation to the depth of the pressure ulcer.

[0012] Furthermore, in the above embodiment, the pressure ulcer level classification is a size-level classification, which is a level classification based on the size of the pressure ulcer, and the classification information may be information indicating the outer circumference of the skin damage area of ​​the pressure ulcer.

[0013] This allows for the accurate acquisition of information indicating the perimeter of the skin damage area of ​​a pressure ulcer. Based on this information indicating the perimeter of the skin damage area of ​​a pressure ulcer, it becomes possible to calculate a level classification for the size of the pressure ulcer, thus enabling accurate calculation of the level classification for the size of the pressure ulcer.

[0014] Furthermore, in the above embodiment, the affected area image is an image in which the affected area and size scale are captured, and the program may further cause the computer to calculate the size of the pressure ulcer based on the size scale and information indicating the outer perimeter of the skin injury area.

[0015] The size scale indicates a predetermined size in the image, allowing for pressure ulcer size measurement regardless of the camera resolution used to capture the image or the distance to the affected area during imaging. This enables accurate calculation of pressure ulcer size based on the size scale.

[0016] Furthermore, in the above embodiment, the affected area image is an image of the pressure ulcer pocket, and the program may further cause the computer to input the affected area image into a fourth learning model that has been trained using fourth training data, which is linked to a learning injured area image and pocket perimeter information indicating the outer circumference of the pressure ulcer pocket in the learning injured area image, and to obtain the pressure ulcer pocket perimeter information from the fourth learning model, and to calculate the size of the pocket based on the size scale, the size of the pressure ulcer, and the pocket perimeter information.

[0017] This makes it possible to calculate the size of pressure ulcer pockets even when the ulcer surface is not exposed. This allows for more accurate classification of pressure ulcer pocket levels.

[0018] An information processing device according to another aspect of the present disclosure is an information processing device for evaluating the state of a pressure ulcer, comprising: a lesion image acquisition unit that acquires a lesion image taken of the affected area; a damage presence / absence acquisition unit that inputs the lesion image to a first learning model trained using first training data linked with a training lesion image different from the lesion image and the presence or absence of skin damage in the training lesion image, and acquires the presence or absence of skin damage in the lesion image; and a classification information acquisition unit that inputs an input image based on the lesion image to a second learning model trained using second training data linked with a training lesion image which is an image of the learning lesion image that includes a lesion in which skin damage has occurred, and classification information for calculating a level classification based on the level of pressure ulcer in the training damage lesion image, and acquires classification information for calculating a level classification of pressure ulcer in the lesion image from the second learning model.

[0019] According to the present invention, a program and an information processing device are provided that can accurately evaluate the condition of pressure ulcers.

[0020] This is a block diagram of the pressure ulcer condition evaluation device according to this embodiment. This diagram illustrates the concept of processing by the injury presence / absence determination model according to this embodiment. This diagram illustrates the concept of processing by the size information output model according to this embodiment. This diagram illustrates the concept of processing by the depth classification information output model according to this embodiment. This diagram illustrates the concept of processing by the pocket perimeter information output model according to this embodiment. This is a flowchart of the pressure ulcer condition evaluation device processing according to this embodiment. This diagram illustrates an example of a screen displayed on the pressure ulcer condition evaluation device according to this embodiment. This is a diagram showing an example of evaluation items according to DESIGN-R2020 classification.

[0021] A preferred embodiment of the present invention will be described with reference to the attached drawings. In each drawing, components denoted by the same reference numerals have the same or similar configuration.

[0022] Figure 1 shows a block diagram of the pressure ulcer condition evaluation device 101 according to this embodiment. The pressure ulcer condition evaluation device 101 includes a display unit 1011, a storage unit 1012, a lesion image acquisition unit 1013, a damage presence / absence acquisition unit 1014, a classification information acquisition unit 1015, a size calculation unit 1016, a pocket size calculation unit 1017, and a level classification calculation unit 1018. Each part of the pressure ulcer condition evaluation device 101 can be realized in an information processing device such as a smartphone, tablet terminal, or personal computer, by having a program stored in a storage device executed by a processor.

[0023] The pressure ulcer condition evaluation device 101 is a device used to evaluate the condition of pressure ulcers. Based on images of the patient, the pressure ulcer condition evaluation device 101 displays to the user the level classification result when DESIGN-R2020 classification is performed on the patient's pressure ulcers. The user makes a final decision on the DESIGN-R2020 classification based on the information displayed on the pressure ulcer condition evaluation device 101.

[0024] In the DESIGN-R2020 classification, there are multiple evaluation items, as shown in Figure 8. For example, for pressure ulcer depth, there are evaluation items d0 to d2 and D3 to DU depending on skin damage, redness, damage to subcutaneous tissue, etc. For pressure ulcer size, there are evaluation items s0 to s12 and S15 depending on the area of ​​skin damage (the area where the ulcer surface of the pressure ulcer is exposed), which is determined by the [long diameter (cm) × short diameter (cm)] of the pressure ulcer (short diameter: the largest diameter perpendicular to the long diameter). Furthermore, for the pocket, which is a wound cavity wider than the skin defect, that is, the area where an ulcer has occurred but is not clearly visible beneath the skin, the area of ​​the pocket is calculated by subtracting the size of the ulcer from the [long diameter (cm) × short diameter (cm)] of the entire circumference of the pocket, including the ulcer surface, and there are evaluation items P0 and P6 to P24.

[0025] Assessments based on the DESIGN-R2020 classification are performed, for example, using visual inspection, palpation, or supplementary data based on observation of pressure ulcers. Assessment results based on the DESIGN-R2020 classification vary depending on the evaluator, and it is particularly difficult for those who are not skilled in assessment to perform an appropriate assessment.

[0026] The pressure ulcer condition evaluation device 101 takes an image of the affected area as input, performs image analysis on the image using a learning model, and generates candidate evaluation items for the DESIGN-R2020 classification. Users of the pressure ulcer condition evaluation device 101 can complete an evaluation based on the DESIGN-R2020 classification based on the information presented by the pressure ulcer condition evaluation device 101.

[0027] The parts of the pressure ulcer condition evaluation device 101 will now be described. The display unit 1011 is the display for the display unit 1011. The storage unit 1012 stores various types of information used in processing by the pressure ulcer condition evaluation device 101. The storage unit 1012 also stores the injury presence / absence determination model 10121, the size information output model 10122, the depth classification information output model 10123, and the pocket perimeter information output model 10124, all of which are used in processing by the pressure ulcer condition evaluation device 101. The storage unit 1012 also stores data generated by the pressure ulcer condition evaluation device 101 and information related to the DESIGN-R2020 classification for pressure ulcer evaluation by the pressure ulcer condition evaluation device 101. The damage detection model 10121, the size information output model 10122, the depth classification information output model 10123, and the pocket perimeter information output model 10124 can use machine learning models including convolutional neural networks (CNNs), such as VGG16, ResNet, or improved versions of these models.

[0028] The injury detection model 10121 (first learning model) is a learning model that has been trained to output whether or not there is skin damage in an image of the affected area, given as input the image of the affected area. The injury detection model 10121 is a learning model that has been trained using training images of the affected area and training data (first training data) that associates the presence or absence of skin damage in the training images of the affected area.

[0029] The size information output model 10122 (second or third learning model) is a learning model that, in response to input images based on affected area images, outputs information indicating the outer perimeter of the skin damage area of ​​a pressure ulcer as classification information in order to calculate the size level classification of pressure ulcers in affected area images. The size information output model 10122 is a learning model that is trained using training affected area images and training data (second or third training data) in which the size information of pressure ulcers in training damaged area images is linked.

[0030] The depth classification information output model 10123 (second learning model) is a learning model that has been trained to output classification information for calculating depth level classification, which is a level classification of the depth of pressure ulcers in the affected area image, in response to input images based on the affected area image. The depth classification information output model 10123 is a learning model that has been trained using training data (second training data) in which training images of injured areas and classification information for calculating depth level classification in the training images of injured areas are linked.

[0031] The pocket perimeter information output model 10124 (fourth learning model) is a learning model that has been trained to output pocket perimeter information indicating the perimeter of a pressure ulcer pocket in response to an input image of the affected area. The pocket perimeter information output model 10124 is a learning model that has been trained using training data (fourth training data) in which training images of the injured area and pocket perimeter information in the training images of the injured area are linked.

[0032] The affected area image acquisition unit 1013 acquires images of the affected area that have been photographed. For example, if the pressure ulcer condition evaluation device 101 is a portable information processing device such as a smartphone or tablet terminal, the affected area image acquisition unit 1013 uses an imaging device (camera) provided in the pressure ulcer condition evaluation device 101 to photograph the affected area and acquire images of the affected area. Alternatively, if the pressure ulcer condition evaluation device 101 is a stationary information processing device such as a personal computer or server, the affected area image acquisition unit 1013 acquires images of the affected area that have been photographed from an imaging device such as a camera or smartphone.

[0033] The damage presence / absence acquisition unit 1014 inputs the affected area image to the damage presence / absence determination model 10121 and acquires from the damage presence / absence determination model 10121 whether or not there is skin damage in the affected area image.

[0034] The classification information acquisition unit 1015 inputs an image of the affected area or an input image based on the image of the affected area to the size information output model 10122, the depth classification information output model 10123, or the pocket perimeter information output model 10124, and acquires classification information from the size information output model 10122, the depth classification information output model 10123, or the pocket perimeter information output model 10124 according to the output of each model. In addition, based on the size information acquired from the size information output model 10122, the classification information acquisition unit 1015 takes an input image extracted from the image of the affected area as input and acquires classification information related to depth from the depth classification information output model 10123. The classification information acquisition unit 1015 performs processing in cooperation with the size calculation unit 1016, the pocket size calculation unit 1017, and the level classification calculation unit 1018, which will be described later.

[0035] The size calculation unit 1016 calculates the size of the pressure ulcer based on the information indicating the outer perimeter of the skin injury area and the size scale (described later) included in the affected area image, which are acquired by the classification information acquisition unit 1015 from the size information output model 10122. Specifically, the size calculation unit 1016 calculates the major and minor axes of the skin injury area (ulcer surface) based on the information indicating the outer perimeter of the skin injury area, and calculates the size (area) of the pressure ulcer as the product of the major and minor axes. By using a size scale captured so as to be included in the affected area image, the size calculation unit 1016 enables measurement of the pressure ulcer size in actual units (cm units) based on the actual size of the size scale, rather than measuring the size of the pressure ulcer in pixels of the image. This makes it possible to measure the size without depending on the shooting conditions (shooting position, shooting distance, etc.) when the affected area image was taken.

[0036] The pocket size calculation unit 1017 calculates the pocket size based on the pocket perimeter information, which indicates the outer circumference of the pocket, obtained by the classification information acquisition unit 1015 from the pocket perimeter information output model 10124, as well as the size scale and the size of the pressure ulcer. Specifically, the pocket size calculation unit 1017 calculates the major and minor diameters of the pocket based on the pocket perimeter information, and calculates the pocket size as the area obtained by subtracting the size of the pressure ulcer from the product of the major and minor diameters.

[0037] The level classification calculation unit 1018 calculates the level classification when DESIGN-R2020 classification is applied to a pressure ulcer, based on the classification information acquired by the classification information acquisition unit 1015, the size of the pressure ulcer calculated by the size calculation unit 1016, or the pocket size calculated by the pocket size calculation unit 1017. The level classification calculation unit 1018 presents the calculated level classification to the user through the display unit 1011.

[0038] The injury presence / absence determination model 10121 will be explained with reference to Figure 2. The injury presence / absence determination model 10121 uses training data in which information such as "skin damage present" or "skin damage absent" is associated with each of several training images of the affected area, including training images TI11, TI12, and TI13. In the example in Figure 2, training image TI11 includes a pressure ulcer BS11, and training image TI12 includes a pressure ulcer BS12, so the information "skin damage present" is associated with each of them. On the other hand, training image TI13 does not contain a pressure ulcer, so the information "skin damage absent" is associated with it.

[0039] The injury detection model 10121 outputs whether or not there is skin damage in the image of the affected area, based on the input image of the affected area. For example, for the input image I11 of the affected area which contains a pressure ulcer BS3, it outputs the information "Skin damage present". On the other hand, for the input image I12 of the affected area which does not contain a pressure ulcer, it outputs the information "No skin damage".

[0040] Referring to Figure 3, the size information output model 10122 will be described. The size information output model 10122 is trained using training data in which information indicating the outer circumference of the pressure ulcer and information indicating the major and minor diameters of the pressure ulcer are associated with training images of affected areas where pressure ulcers exist.

[0041] In the example of FIG. 3, for the learning affected - part image TI21 including the pressure ulcer BS21, information indicating its outer periphery R21 and information indicating the major axis MA21 and minor axis MI21 of the pressure ulcer BS21 are associated. Also, the learning affected - part image TI21 may include a size scale SS having a predetermined size. The size scale SS is used for calculating the size of the pressure ulcer by the size calculation unit 1016 and calculating the pocket size of the pressure ulcer by the pocket size calculation unit 1017, which will be described later. Similarly, for the learning affected - part image TI22 including the pressure ulcer BS22, information indicating the outer periphery R22 and information indicating the major axis MA22 and minor axis MI22 of the pressure ulcer BS22 are associated. Similarly, for the learning affected - part image TI23 including the pressure ulcer BS23, information indicating the outer periphery R23 and information indicating the major axis MA23 and minor axis MI23 of the pressure ulcer BS23 are associated.

[0042] The size information output model 10122 outputs information indicating the outer periphery of the skin damage range of the pressure ulcer and information indicating the major axis and minor axis of the pressure ulcer in response to the input of the affected - part image. In the example of FIG. 3, the size information output model 10122 outputs, as classification information, information indicating the outer periphery R3 of the pressure ulcer BS3 in the affected - part image I11, the major axis MA3, and the minor axis MI3 in response to the input of the affected - part image I11.

[0043] Referring to FIG. 4, the depth classification information output model 10123 will be described. The depth classification information output model 10123 outputs classification information (depth classification information) for calculating a depth level classification, which is a level classification for the depth of the pressure ulcer in the affected - part image, in response to the input of an input image based on the affected - part image.

[0044] The depth classification information output model 10123 is trained using teacher data in which the depth level classification of a bedsore is associated with a learning target lesion image cut out based on the size of the bedsore in the lesion image. In the example of FIG. 3, for the learning target lesion image TI31 obtained by cutting out the learning target lesion image TI21 with the frame C21 generated based on the major axis MA21 and the minor axis MI21 of the bedsore BS21, it is associated that the depth level classification of the bedsore is D3. Also, for the learning target lesion image TI32 obtained by cutting out the learning target lesion image TI22 with the frame C22 generated based on the major axis MA22 and the minor axis MI22 of the bedsore BS22, it is associated that the depth level classification of the bedsore is d2. Further, for the learning target lesion image TI33 obtained by cutting out the learning target lesion image TI23 with the frame C23 generated based on the major axis MA23 and the minor axis MI23 of the bedsore BS23, it is associated that the depth level classification of the bedsore is d1.

[0045] The depth classification information output model 10123 outputs depth classification information for calculating a depth level classification for an input of an input image extracted based on information indicating the outer periphery of a bedsore output by the size information output model 10122. The depth level classification information is a probability indicating whether the depth of the bedsore is classified into any of the depth level classifications. In the example of FIG. 4, for an input of a lesion image I11 including the bedsore BS3, the input image I41 obtained by extracting the lesion image I11 with the frame C3 based on the information of the outer periphery R3, the major axis MA3, and the minor axis MI3 output by the size information output model 10122 is input to the depth classification information output model 10123.

[0046] The depth classification information output model 10123 outputs depth classification information of the bedsore BS3. For example, in the example of FIG. 4, an example is shown in which the depth classification information is obtained as output such as d0: 0.1%, d1: 1.02%, d2: 12.47%, D3: 85.02%, D4: 1.39%, etc.

[0047] Referring to Figure 5, the pocket perimeter information output model 10124 will be described. The pocket perimeter information output model 10124 is trained using training data that associates information indicating the perimeter of the pocket and information indicating the major and minor axes of the pocket in training images of the affected area where a pressure ulcer and pocket exist. Here, the pocket is indicated in the training image of the affected area or the affected area image by marking information indicating the pocket, for example, through visual inspection or palpation.

[0048] In the example shown in Figure 5, information indicating the outer circumference PR21 of pocket P21 and information indicating the major axis PMA21 and minor axis PMI21 of pocket P21 are associated with the learning image of the affected area TI21, which includes the pressure ulcer BS21 and pocket P21. Similarly, information indicating the outer circumference PR22 of pocket P22 and information indicating the major axis PMA22 and minor axis PMI22 of pocket P22 are associated with the learning image of the affected area TI22, which includes the pressure ulcer BS22 and pocket P22. Furthermore, information indicating the outer circumference PR23 of pocket P23 and information indicating the major axis PMA23 and minor axis PMI23 of pocket P23 are associated with the learning image of the affected area TI23, which includes the pressure ulcer BS23 and pocket P23.

[0049] Pockets P21 and P23 are partial pockets that do not encompass the entire pressure ulcer, while pocket P22 is a full-circumference pocket that encompasses the entire pressure ulcer. When the pressure ulcer is a partial pocket, the information indicating the outer circumference of the pocket shows the area around the pocket and the pressure ulcer, excluding the part where the pocket and the pressure ulcer are in contact. When the pressure ulcer is a full-circumference pocket, the information indicating the outer circumference of the pocket shows the outer circumference of the pocket.

[0050] The pocket perimeter information output model 10124 outputs information indicating the perimeter of the pocket and information indicating the major and minor diameters of the pocket in response to an input image of the affected area. In the example shown in Figure 5, the pocket perimeter information output model 10124 outputs information indicating the perimeter PR3 of the pocket in the affected area image I11, which includes the pressure ulcer BS3 and pocket P3, and information indicating the major diameter PMA3 and minor diameter PMI3 as classification information in response to an input image I11 of the affected area.

[0051] In the pressure ulcer condition evaluation device 101, the injury presence / absence determination model 10121 first determines whether or not a pressure ulcer is present in the image. Subsequently, images in which a pressure ulcer is determined to be present are input to the size information output model 10122, the depth classification information output model 10123, or the pocket perimeter information output model 10124.

[0052] In this way, by first determining the presence or absence of pressure ulcers and inputting only images containing pressure ulcers into subsequent models, the size information output model 10122, the depth classification information output model 10123, or the pocket perimeter information output model 10124 no longer need to detect the presence or absence of pressure ulcers. This allows each model to be trained to be more specialized in pressure ulcers themselves. As a result, the accuracy of pressure ulcer condition evaluation in the pressure ulcer condition evaluation device 101 is improved.

[0053] Furthermore, the pressure ulcer condition evaluation device 101 eliminates the need to prepare separate learning models for each classification item of the DESIGN-R2020 classification (for example, a model for evaluating d1, a model for evaluating d2, or a model for evaluating d3), thus enabling faster processing of images of the affected area.

[0054] Referring to Figure 6, the processing by the pressure ulcer condition evaluation device 101 will be explained. In step S601, the affected area image acquisition unit 1013 acquires an image of the affected area. In step S602, the injury presence / absence acquisition unit 1014 inputs the affected area image into the injury presence / absence determination model 10121.

[0055] In step S603, the damage presence / absence acquisition unit 1014 acquires the presence or absence of skin damage in the affected area image from the damage presence / absence determination model 10121 and determines whether or not there is skin damage.

[0056] If a negative result is obtained in step S603, that is, if there is no skin damage in the affected area image, the process is terminated.

[0057] If a positive determination is made in step S603, that is, if there is skin damage in the affected area image, in step S604, the classification information acquisition unit 1015 inputs the affected area image into the size information output model 10122.

[0058] In step S605, the classification information acquisition unit 1015 acquires the outer circumference information of the pressure ulcer from the size information output model 10122.

[0059] Subsequently, processes are carried out to calculate the depth level classification of pressure ulcers, the size level classification of pressure ulcers, and the pocket level classification of pressure ulcers. These processes may be carried out in parallel or sequentially.

[0060] First, let's explain how to calculate the depth level classification. In step S606, the classification information acquisition unit 1015 extracts an input image from the affected area image and inputs the input image into the depth classification information output model 10123.

[0061] In step S607, the classification information acquisition unit 1015 acquires classification information related to depth from the depth classification information output model 10123.

[0062] In step S608, the level classification calculation unit 1018 calculates the depth level classification based on the classification information relating to depth. For example, if the classification information relating to depth is d0:0.1%, d1:1.02%, d2:12.47%, D3:85.02%, D4:1.39%, the level classification calculation unit 1018 calculates the depth level classification with the highest probability as the depth level classification. In this example, the level classification calculation unit 1018 calculates the depth level classification as D3.

[0063] Next, the calculation of size level classification will be explained. In step S609, the size calculation unit 1016 calculates the size of the pressure ulcer based on the outer circumference information of the pressure ulcer and the size scale included in the affected area image. Specifically, the size calculation unit 1016 obtains information on the major and minor diameters based on the outer circumference information of the pressure ulcer and calculates the lengths of the major and minor diameters based on the size scale. The size calculation unit 1016 calculates the size of the pressure ulcer as the product of the calculated major diameter value and minor diameter value.

[0064] In step S610, the level classification calculation unit 1018 calculates the size level classification of the pressure ulcer based on the size of the pressure ulcer calculated by the size calculation unit 1016. The size calculation unit 1016 calculates the size level classification of the pressure ulcer based on the size of the pressure ulcer and the evaluation items in the DESIGN-R2020 classification.

[0065] Finally, the calculation of pocket level classification will be explained. In step S611, the pocket size calculation unit 1017 inputs the affected area image to the pocket perimeter information output model 10124.

[0066] In step S612, the pocket size calculation unit 1017 acquires pocket outer circumference information from the pocket outer circumference information output model 10124.

[0067] In step S613, the pocket size calculation unit 1017 calculates the size of the pocket based on the pocket perimeter information and the size scale included in the affected area image. Specifically, the pocket size calculation unit 1017 obtains information on the major and minor diameters of the pocket based on the pocket perimeter information and calculates the lengths of the major and minor diameters based on the size scale. The pocket size calculation unit 1017 calculates the size of the pocket as the product of the calculated major and minor diameter values. The pocket size is calculated by subtracting the size of the pressure ulcer from the product of the major and minor diameters of the entire circumference of the pocket, including the ulcer surface of the pressure ulcer.

[0068] In step S614, the level classification calculation unit 1018 calculates the pocket size level classification based on the pocket size calculated by the pocket size calculation unit 1017. The pocket size calculation unit 1017 calculates the pocket size level classification based on the pocket size and the evaluation items in the DESIGN-R2020 classification.

[0069] In step S615, the level classification calculation unit 1018 displays the pressure ulcer level classification in the affected area image to the user via the display unit 1011. Even if a negative determination is made in step S603 (i.e., the affected area image does not contain a pressure ulcer), the level classification calculation unit 1018 displays the pressure ulcer level classification, such as s0 or d0, to the user.

[0070] Referring to Figure 7, the screen 701 displayed to the user by the pressure ulcer condition evaluation device 101 will be explained. In the example in Figure 7, the pressure ulcer condition evaluation device 101 is a smartphone, and the pressure ulcer condition evaluation device 101 has captured an image of the affected area of ​​patient P having a pressure ulcer BS.

[0071] Screen 701 displays images of the pressure ulcer BS3, pocket P3, and size scale SS captured by the pressure ulcer condition evaluation device 101. In addition, the major axis MA3 and minor axis MI3, based on information indicating the outer circumference R3 of the pressure ulcer acquired by the classification information acquisition unit 1015, are superimposed on the image and displayed. Furthermore, the major axis PMA3 and minor axis PMI3, based on information indicating the outer circumference of the pocket P3 acquired by the pocket size calculation unit 1017, are also superimposed on the image and displayed.

[0072] Furthermore, the level classification display area 7011 displays the level classification finally calculated by the level classification calculation unit 1018. In the example shown in Figure 7, an example is shown where D=D3 is calculated and displayed as the depth level classification, S=s8 as the size level classification, and P=P9 as the pocket level classification.

[0073] Furthermore, the level classification display area 7011 may also display information that forms the basis for calculating the level classification. In the example in Figure 7, the classification information (probability) for the depth level classification is displayed as follows: "(D) d0: 0.1% d1: 1.02% d2: 12.47% D3: 85.02% D4: 1.39%..." Also, for the size level classification, "(S) s0: 0.01% Length (major axis): 4.92 cm Width (minor axis): 3.53 cm S area: 17.4 cm ２ The following information regarding the size is displayed: "(P) Length (short axis): 4.32 cm, Width (long axis): 6.53 cm, S+P area: 28.2 cm ２ , P area: 10.8cm ２ Information about its pocket size is displayed.

[0074] By presenting users with both the level classification and the information supporting its calculation, users can ultimately evaluate pressure ulcers based on the DESIGN-R2020 classification by reviewing both the evaluation candidates and the reasons supporting them. This allows users to more accurately evaluate pressure ulcers based on the DESIGN-R2020 classification.

[0075] This embodiment has been described above. In the case of the pressure ulcer condition evaluation device 101, depth, size, and pocket were used as examples to explain the DESIGN-R2020 classification, but the method of performing evaluations using each model after determining the presence or absence of a pressure ulcer by the pressure ulcer condition evaluation device 101 can also be applied to the evaluation of "Inflammation / Infection," "Granulation," and "Necrotic tissue."

[0076] The embodiments described above are provided to facilitate understanding of the present invention and are not intended to limit its interpretation. The elements, arrangement, conditions, shape, and size of each embodiment are not limited to those exemplified and can be modified as appropriate. Furthermore, it is possible to partially substitute or combine the configurations shown in different embodiments.

[0077] 101... Pressure ulcer condition evaluation device, 1013... Affected area image acquisition unit, 1014... Injury presence / absence acquisition unit, 1015... Classification information acquisition unit, 1016... Size calculation unit, 1017... Pocket size calculation unit, 1018... Level classification calculation unit, 10121... Injury presence / absence determination model, 10122... Size information output model, 10123... Depth classification information output model, 10124... Pocket perimeter information output model

Claims

1. A program that causes a computer to evaluate the condition of a pressure ulcer, the program to perform the following actions: acquire an image of the affected area taken from the affected area; input the image of the affected area into a first learning model, which has been trained using first training data that links a different training image of the affected area with the presence or absence of skin damage in the training image of the affected area, and acquire the presence or absence of skin damage in the image of the affected area; and input an input image based on the image of the affected area into a second learning model, which has been trained using second training data that links a training image of the affected area that includes the area with skin damage, and classification information for calculating a level classification based on the level of the pressure ulcer in the training image of the damaged area, and acquire classification information for calculating a level classification of the pressure ulcer in the image of the affected area from the second learning model.

2. The program according to claim 1, wherein the level classification of the pressure ulcer is a depth level classification which is a level classification of the depth of the pressure ulcer, and the classification information is a probability indicating whether the depth of the pressure ulcer is classified into any of the depth level classifications.

3. The program according to claim 2, wherein the computer is further instructed to input the injured area image to a third learning model trained using third training data linked to the learning injured area image and the size information of the pressure ulcer in the learning injured area image, and to obtain the size information of the pressure ulcer from the third learning model, and to obtain the classification information from the second learning model, the input image extracted from the injured area image based on the size information is used as input to obtain the classification information from the second learning model.

4. The program according to claim 1, wherein the level classification of the pressure ulcer is a size-level classification which is a level classification relative to the size of the pressure ulcer, and the classification information is information indicating the outer circumference of the skin damage area of ​​the pressure ulcer.

5. A program according to claim 4, wherein the affected area image is an image of the affected area and a size scale, and the program further causes the computer to calculate the size of the pressure ulcer based on the size scale and information indicating the outer circumference of the skin injury area.

6. A program according to claim 5, wherein the affected area image is an image of the pocket of the pressure ulcer, and the program further causes the computer to input the affected area image into a fourth learning model which has been trained using fourth training data in which the learning injured area image and pocket perimeter information indicating the outer circumference of the pocket of the pressure ulcer in the learning injured area image are linked, to obtain the pocket perimeter information of the pressure ulcer from the fourth learning model, and to calculate the size of the pocket based on the size scale, the size of the pressure ulcer, and the pocket perimeter information.

7. An information processing device for evaluating the state of a pressure ulcer, comprising: a lesion image acquisition unit that acquires a lesion image taken of the affected area; a damage presence / absence acquisition unit that inputs the lesion image to a first learning model trained using first training data linked with a training lesion image different from the lesion image and the presence or absence of skin damage in the training lesion image, and acquires the presence or absence of skin damage in the lesion image; and a classification information acquisition unit that inputs an input image based on the lesion image to a second learning model trained using second training data linked with a training lesion image which is an image of the training lesion image that includes the lesion in which the skin damage has occurred, and classification information for calculating a level classification based on the level of the pressure ulcer in the training damage lesion image, and acquires classification information for calculating the level classification of the pressure ulcer in the lesion image from the second learning model.

Images