WO2022149566A1

WO2022149566A1 - Remote system for determining risk of skin diseases and remote method for determining risk of skin diseases

Info

Publication number: WO2022149566A1
Application number: PCT/JP2022/000025
Authority: WO
Inventors: 公一朗木島
Original assignee: ピンポイントフォトニクス株式会社
Priority date: 2021-01-06
Filing date: 2022-01-04
Publication date: 2022-07-14
Also published as: JPWO2022149566A1

Abstract

Risk determining systems that perform machine learning for an image capturing a lentigo need preparation through use of a lot of learning data in consideration of variations in illumination amount because a feature amount of the image such as morphology or texture is influenced by variations in illumination conditions. In this remote system for determining the risk of skin diseases and this remote method for determining the risk of skin diseases, prior to analyzing texture information and morphology information, brightness information is removed from an image capturing the skin through a boundary detection processing step for outputting information including a tone value, thereby enabling removal of an influence of variations in illumination conditions from the analysis results of the text information and the morphology information. This reduces the need of preparing learning data corresponding to various illumination conditions.

Description

Remote skin disease risk determination system and remote skin disease risk determination method

Melanoma (melanoma) is a type of skin cancer and is a skin disease that has been increasing in Japan in recent years. Melanoma is present on the surface of the skin. That is, the melanoma exists in a position where it can be visually recognized without using a device such as an endoscope or an X-ray device possessed only by a medical institution. Therefore, as shown in Patent Document 1, the discolored part of the skin is photographed by a portable device such as a smartphone, and the photographed image is transferred to a remote server via an internet line to determine the risk of melanoma. A remote skin disease risk determination system has been proposed.

As shown in FIG. 1, in the remote skin disease risk determination system 10 shown in Patent Document 1, for example, a user who may be a patient takes a picture of a mole 12 existing on the skin surface 9 of a foot by a smartphone 11, and an internet line is used. Transfer to the server 20 owned by the service provider via 19. After confirming the patient information by the image reception reception tool 21, the service provider confirms by the determination tool 22 whether or not this image is a target image for determining the risk of skin diseases such as moles and melanoma. Then, when it is determined that the image is for determining the risk of skin disease, the image data such as the texture, morphology, and color information of the mole 12 is extracted from the image by the feature parameter extraction tool 23. Next, this image data is classified by the image classification tool 24 and the risk determination tool 25 that have been machine-learned by the case learning data, and then the determination result is given to the user.

This remote skin disease risk determination system has the advantage that the burden on the user is small because the user does not have to go to the hospital. Also, on the medical institution side, there is an advantage that the burden on the medical staff is small because the screening is performed by the machine-learned device.
However, the image of the mole taken by a potential patient is not always taken under the same lighting conditions as the image used for machine learning. Image data such as texture, morphology, and color information of the hokuro 12 used in general machine learning image judgment systems vary depending on the lighting conditions, so it is necessary to prepare training data corresponding to many lighting conditions. There are drawbacks. Therefore, the present invention relates to a remote skin disease risk determination system and a remote skin disease risk determination method that can extract image data such as texture, morphology, and color information of mole 12 without variation even when the lighting conditions vary. Further, when a user takes a picture of a mole with a smartphone, blurring or the like is likely to occur if there is no member that holds a relative position between the position of the smartphone and the mole. In recent years, the number of pixels of cameras installed in smartphones has been increasing, so it is difficult for the photographer to notice blurring that cannot be confirmed when the confirmation of the captured image is displayed at a low magnification, and the image in which this blurring exists. There is a high possibility that the photographer will send the image to the server. An image with blur is determined as an ineligible image by the determination tool 22, so that the determination result is not affected. Therefore, the load on the user increases.

The variation in the color information of the subject photographed by an image photographing means such as a camera due to the illumination state can be corrected by a method generally used as a white balance correction. However, a method for correcting variations in texture and morphology image data due to lighting conditions is not generally used.
An example will be described in which the image data such as the texture and morphology of the mole 12 used in a general image determination system have variations depending on the lighting state. FIG. 2 shows a photograph in which the mole is arranged almost in the center. Then, FIGS. 3 (a) to 3 (d) are photographs in which the luminance information of the photograph shown in FIG. 2 is changed as an example of images taken under different lighting conditions. FIG. 3A is a photograph in which the right end of the photograph has the same luminance as the photograph shown in FIG. 2, but the luminance information is changed so that the left edge of the photograph has a luminance of 75%. Specifically, as shown in FIG. 4, the upper left coordinate of the image is (0,0), the upper right coordinate is (X, 0), the lower left coordinate is (0, Y), and the lower right coordinate is (X). , Y), the luminance I (i, j) of each coordinate (i, j) in FIG. 4A is relative to the luminance P (i, j) of the image shown in FIG. 2 as shown in Equation 1. It is an image having a brightness such that the brightness changes linearly according to the coordinates x after being multiplied by 0.75. Similarly, FIG. 3B is a photograph in which the right end of the photograph has the same luminance as the photograph shown in FIG. 2, but the luminance information is changed so that the left edge of the photograph has a luminance of 50%. As shown in Equation 2, the image is multiplied by 0.5 with respect to the brightness P (i, j) of the image shown in FIG. 2, and then the brightness is set so that the brightness changes linearly according to the coordinates x. Is. Therefore, the image shown in FIG. 3A is an image in which the illumination state has an inclination of 25%, and the image shown in FIG. 3B is an image in which the illumination state has an inclination of 50%.

The image shown in FIG. 3C is an image in which the lighting state is uniformly reduced by 25%, and the image shown in FIG. 3D is an image in which the lighting state is uniformly reduced by 50%. The image shown in FIG. 3 (c) and the image shown in FIG. 3 (d) depend on the coordinate position with respect to the brightness P (i, j) of the image shown in FIG. 2 as shown in

Equations

3 and 4. No, 0.75x and 0.5x images.

FIG. 5 shows the results of texture analysis of the three areas shown by the white lines in the image shown in FIG. 2 and the four images shown in FIG. The caption described as original in FIG. 5 is the result of texture analysis of the portions shown in areas 1 to 3 of the image of FIG. 2, and (a) gradation (25%), (b) gradation (50%), ( c) The captions described as reduction (25%) and (d) reduction (50%) were subjected to texture analysis of the parts shown in areas 1 to 3 of the images shown in FIGS. 3 (a) to 3 (d), respectively. The result. A general entropy was used as the texture analysis method.

From this result, it is shown that the image data of the texture of the mole 12 varies due to the variation of the lighting state, and it is necessary to prepare the learning data corresponding to many lighting conditions. .. Further, in the numerical value of morphology, the line width of the boundary changes due to the variation of the illumination state, which affects the analysis value of morphology.

In order to accurately analyze the texture information and the morphology information, it is desirable to analyze the texture information and the morphology information by a method that does not depend on the lighting method. Therefore, we attempted to study the boundary detection processing step before the feature parameter extraction step, which is the analysis step of texture information and morphology information.
FIG. 6 shows an image feature analysis method using the phase stretch transform, which is the boundary detection processing step shown in Patent Document 2 and Non-Patent Document 1, on the images shown in FIGS. 2 and 3 (a) to 3 (d). After application, the results of texture analysis of the three areas shown by the white lines are shown. The phase stretch transform is a method of performing edge detection by a non-linear detection method that combines phase information and derivative information of phase information in the phase information of an image obtained by Fourier transforming the image information. Since the edge information detected by this method is not binarized information but information having a gradation value, it is also possible to exclude the luminance information which is the background information by performing this edge detection. be. As the texture analysis method, the same general entropy as the result in FIG. 5 was used.

From the results shown in FIG. 6, the luminance information is removed by the boundary detection processing step that outputs the information having the gradation value before the texture information and the morphology information are analyzed, so that the texture analysis result varies in the lighting state. It turns out that it is not affected by. The fact that the texture analysis results are not affected by the variation in lighting conditions indicates that even if the lighting conditions of the image taken by the user vary, the texture can be analyzed without being affected by the variation. Therefore, it is not necessary to prepare training data corresponding to many lighting conditions.

US Pat. No. 5,543,519 US Pat. No. 10,275,891

The present invention has been made in consideration of the above points, and by allowing variations in the lighting conditions of the shooting environment, variations in image data such as the texture and morphology of the hokuro become large, and therefore, many lighting conditions can be obtained. It provides a solution to the problem that it is necessary to prepare the corresponding learning data.

In order to solve this problem, in the remote skin disease risk determination system and the remote skin disease risk determination method of the present invention, a boundary detection processing step of outputting information having a gradation value is used before analyzing texture information and morphology information. By removing the brightness information from the image of the skin taken, the effect of variation in lighting conditions is excluded from the analysis results of the texture information and morphology information.
Further, when the user takes an image of the skin with the smartphone, by using a member that holds the relative position between the position of the smartphone and the mole, blurring and the like are less likely to occur. Furthermore, a member on which a scale is formed is placed on a member that holds the relative position between the position of the smartphone and the mole, and the image is photographed. By making the correction of, it is possible to accurately acquire the size information of the mole, and it is possible to use the image without distortion in the classification step and the risk judgment step. A member for correcting color information is arranged on a member that holds the relative position between the position of the smartphone and the mole, and by taking a picture of this member, the color information of the lighting is acquired and the color is corrected. Enables.

Since the remote skin disease risk determination system and the remote skin disease risk determination method of the present invention can exclude the influence of variations in lighting conditions from the analysis results of texture information and morphology information, learning data corresponding to many lighting conditions can be excluded. The need to prepare is reduced.
Further, the remote skin disease risk determination system and the remote skin disease risk determination method of the present invention illuminate by arranging a member that holds a relative position between the position of the smartphone and the mole when taking an image of the skin with the smartphone. Even if the state changes, the influence of the variation in the lighting state can be excluded from the analysis results of the texture information and the morphology information. Therefore, it is possible to arrange a member that holds the relative position between the position of the smartphone and the mole, and it becomes easy to take an image without blurring.
Furthermore, a member on which a scale is formed is placed on a member that holds the relative position between the position of the smartphone and the mole, and the image is photographed. By making the correction of, it is possible to accurately acquire the size information of the mole, and it is possible to use the image without distortion in the classification step and the risk judgment step.

It is a schematic block diagram of the remote skin disease risk determination system as a prior art. This is an example of a mole image used in a remote skin disease risk determination system. This is an example of a mole image with variations in lighting conditions used in a remote skin disease risk determination system. It is a figure which shows the method of making the image of the mole with the variation of lighting conditions to be used for the remote skin disease risk determination system. It is a figure of the experimental result which shows that the texture analysis in the remote skin disease risk determination system as a prior art varies depending on the variation of lighting conditions. It is a figure of the experimental result which shows that the influence of the variation of a lighting state is excluded from the analysis result of the texture information and the morphology information by performing the boundary detection processing step before the texture analysis step. It is a schematic block diagram of the remote skin disease risk determination system of this invention. It is a schematic block diagram of the member which holds the relative position of a smartphone and a mole in the remote skin disease risk determination system of this invention. It is a schematic block diagram which attached the member which holds the relative position between the position of the smartphone and the mole in the remote skin disease risk determination system of this invention to the smartphone. It is a schematic block diagram of the member which holds the position of the 2nd smartphone and the relative position of a mole in the remote skin disease risk determination system of this invention. It is a schematic block diagram which attached the member which holds the position of the 2nd smartphone and the relative position of a mole in the remote skin disease risk determination system of this invention to a smartphone. It is a schematic diagram which shows the method of correcting the color information by using the member which holds the relative position of the 2nd smartphone and the mole in the remote skin disease risk determination system of this invention. It is sectional drawing of the scale member which fits into the member which holds the position of a smartphone and the relative position of a mole in the remote skin disease risk determination system of this invention. It is a schematic block diagram in the case of taking a picture of the scale member which fits into the member which holds the relative position between the position of the smartphone and the mole in the remote skin disease risk determination system of this invention with a smartphone. It is a block diagram of the scale member which fits into the member which holds the position of a smartphone and the relative position of a mole in the remote skin disease risk determination system of this invention. The remote skin disease risk determination system having a step of excluding distortion characteristics of a camera by a scale of a member that fits a member that holds a relative position between the position of a smartphone and a mole of the present invention and a step of correcting color information of lighting. It is a schematic block diagram. It is a schematic process diagram which shows the process of the application which a user performs through a smartphone in the remote skin disease risk determination system of this invention. It is a schematic process diagram which shows the process which the user acquires the condition which corrects the distortion of an image in the remote skin disease risk determination system of this invention using the application of a smartphone.

FIG. 7 shows a schematic configuration diagram of the remote skin disease risk determination system of the present invention.
In the remote skin disease risk determination system 30 of the present invention, for example, a mole 12 formed on the skin surface 9 of a foot is photographed by a smartphone 11 which is a portable device having an image capturing function and a data transfer function, and an internet line 19 is taken. The data is transferred to the server 31 owned by the service provider. After confirming the patient information by the image reception reception tool 21, the service provider confirms by the determination tool 22 whether or not this image is a target image for determining the risk of skin diseases such as moles and melanoma. Then, when it was determined that the image was used to determine the risk of skin disease, the image feature analysis method using the phase stretch transform, which is the boundary detection processing step shown in Patent Document 2 and Non-Patent Document 1, was applied. The background brightness information is excluded by the boundary / texture extraction tool 29 that outputs information having a gradation value. Next, the image data such as the texture and morphology of the mole 12 is extracted from the image by the feature parameter extraction tool 23. After that, this image data is classified by the image classification tool 24 and the risk determination tool 25 that have been machine-learned by the case learning data, and then the determination result is given.

In the remote skin disease risk determination system of the present invention, as shown in the experimental results of FIGS. 5 and 6, the feature parameter extraction tool 23 is applied after the background brightness information is excluded by the boundary / texture extraction tool 29. By extracting image data such as texture and morphology of the hokuro 12 that does not depend on the lighting conditions from the image, it is possible to exclude variations in image data such as texture and morphology due to variations in lighting conditions in the captured image. Therefore, even if the lighting state of the image taken by the smartphone 11 is non-uniform, or the image is dark as a whole, the texture, morphology, etc. similar to those of the image taken under the optimum lighting state can be obtained. Image data can be extracted.
Furthermore, since it is not necessary to use learning data considering variations in lighting conditions for the data machine-learned by the case learning data used in the image classification tool 24 and the risk determination tool 25, less learning data than in the conventional example. Can be used.

In the above description of the invention, an example using the boundary / texture extraction tool 29 to which the image feature analysis method using the phase stretch transform, which is the boundary detection processing step shown in Patent Document 2 and Non-Patent Document 1, is applied. As shown, it is also possible to use another method of boundary detection as long as it has a function of extracting edge information having a gradation value.

FIG. 8 shows a schematic configuration diagram of the member 40 that holds the relative position between the position of the smartphone and the mole. This member is a macro that makes it easy to align the focal position of the camera of the smartphone 11 with the spacer member 41 that keeps the distance between the smartphone and the mole, which is a portable device having an image capturing function and a data transfer function, constant. The structure is such that the photographing lens 42 is fixed. Further, a clip member 43 for holding a state in which the macro photography lens 42 is positioned substantially in front of the camera of the smartphone 11 is also provided. The clip member has a structure in which the two

clip holders

44 and 45 can be opened and closed from the fulcrum 46, and the smartphone 11 can be sandwiched by the spring force of the spring 47.

FIG. 9 shows a configuration diagram in which the mole 12 is photographed in a form in which the member 40 that holds the relative position between the position of the smartphone and the mole is fixed to the smartphone 11 by using the clip member 43. By taking an image of the mole 12 in a state where the member 40 that holds the relative position between the position of the smartphone and the mole is fixed to the smartphone 11, it is possible to prevent blurring or the like from occurring in the taken image. The spacer member 41 of the member 40 that holds the relative position between the position of the smartphone and the mole is formed of a transparent member that transmits external light. Since the spacer member 41 is made of a transparent member, the image of the mole 12 can be taken by using outside light such as indoor lighting without having a lighting mechanism.

It is desirable that the spacer member 41 has a cylindrical shape or a conical shape without apex portions as shown in FIGS. 8 and 9, and the inner surface 41a thereof is a sand-treated surface. Since the inner surface 41a is a surface treated with sand, it is possible to prevent the shadow of the illumination from being clearly reflected in the image. Since the sand-treated surface is arranged on the inner surface 41a side instead of the outer peripheral surface of the spacer member 41, the risk of the photographer or the like touching the sand-treated surface is reduced, and oil or the like adheres to the sand-treated surface. It is possible to reduce the risk of deterioration of the diffused reflection characteristics of the surface.

When the member 40 that holds the relative position between the position of the smartphone and the mole is used, the illumination state of the image of the mole 12 is poor as compared with the case where the member 40 that holds the relative position between the position of the smartphone and the mole is not used. Although there is a high possibility that the image will be uniform, in the present invention, by excluding the background brightness information by the boundary / texture extraction tool 29 that outputs information having a gradation value, the texture due to the variation in the illumination state in the captured image, Since variations in image data such as morphology can be excluded, it is possible to use the member 40 that holds the relative position between the position of the smartphone and the mole.

FIG. 10 shows a schematic configuration diagram of a second member 39 that holds a relative position between the position of the smartphone and the mole. The member 39 has a structure in which a color correction sheet 48 having a white color on the member 40 shown in FIG. 8 is fixed to a spacer member 41, and the color correction sheet 48 has an opening 48a so that a mole can be photographed. When this is done, the color correction sheet 48 is also arranged so as to be photographed in a part of the field of view. Further, as in the case of the member 40, the clip member 43 is used to fix the member 39 that holds the relative position between the position of the smartphone and the mole to the smartphone 11, and the configuration diagram of the case where the mole 12 is photographed is shown in FIG. Shown in.

FIG. 12 shows an example of an image of the mole 12 taken by using the second member 39 that holds the relative position between the position of the smartphone and the mole. In FIG. 12, the image 49 of the color correction sheet 48 is photographed around the mole 12. Next, a method of correcting the color information of the illumination of the shooting environment from the information of the image 49 of the color correction sheet will be described. Luminance information of a plurality of arbitrary points of the image 49 of the color correction sheet, for example, three points of 49a, 49b, and 49c is analyzed, and the red, green, and blue information of those points are (100, 110, 60), respectively. Assuming (150,165,90) and (60,66,36), since the color correction sheet 48 is white, the photographed lighting environment has the strongest green and the weakest blue. It can be seen from this information, and in the captured image shown in FIG. 12, the green luminance information is left as it is, and the red luminance information is multiplied by 1.1 in the three luminance information of 49a, 49b, and 49c to be blue. By multiplying the luminance information by 110/60, it can be obtained as (110, 110, 110), (165, 165, 165), (66, 66, 66), respectively. The same correction is applied to the image portion of the hokuro 12, specifically, the green luminance information is left as it is, and the red luminance information is multiplied by 1.1 in the three luminance information of 49a, 49b, and 49c to obtain the blue luminance information. By performing a correction of 110/60 times, it is possible to reproduce the color information when the image is taken under uniform illumination with a uniform balance of red, green, and blue.

FIG. 13 shows a cross-sectional structural view of a member 39 that holds the relative position between the position of the smartphone and the mole and the scale member 51 that fits into the spacer member 41 of the member 40. FIG. 14 shows a configuration diagram of the scale member 51 attached to the member 40 that holds the relative position between the position of the smartphone attached to the smartphone 11 and the mole. The scale member 51 has a configuration in which the surface 51a is arranged at the position of the skin when an image of a mole is taken by the smartphone 11 using the spacer member 41, and the scale member 51 has a scale arranged on the surface 51a. There is. FIG. 15 shows an example of a scale formed on the surface 51a of the scale member 51. The scale includes, for example, a concentric scale pattern 52 and

linear scale patterns

53 and 54 orthogonal to each other. By photographing the

scale patterns

52, 53, 54 with the smartphone 11, it is possible to know the distortion information of the image of the image of the mole 12 located on the skin surface 9. That is, when the

linear scale patterns

53 and 54 are distorted in the image in which the scale pattern is photographed, it can be seen that the same distortion is added to the image in the image in which the mole 12 is photographed. Therefore, if the

scale patterns

52, 53, 54 are distorted in the image obtained by capturing the scale pattern, the distortion correction is performed so that the distorted

scale patterns

52, 53, 54 become the actual scale pattern shown in FIG. That is, it is possible to generate an image without distortion by performing distortion correction. Further, it is possible to correctly know the shooting magnification by shooting the

scale patterns

52, 53, 54 with the smartphone 11.

In the remote skin disease risk determination system of the present invention, by transferring the photograph of the scale 51 together with the image of the mole 12 from the smartphone to the server, the risk can be determined in the image without distortion, and the scale information is also a risk. It can be used for judgment.

FIG. 16 shows an image of the mole 12 taken by using the member 39 or the member 40 that holds the relative position between the position of the smartphone and the mole, and further, the member 39 or the member 40 that holds the relative position between the position of the smartphone and the mole. The schematic block diagram of the remote skin disease risk determination system 60 which uses the image which image | photographed the scale member 51 with, is shown. In the remote skin disease risk determination system 60 of the present invention, for example, a mole formed on the skin surface 9 of a foot is photographed by a smartphone 11 using a member 40 that holds a relative position between the position of the smartphone and the mole, and an internet line is used. Transfer to the server 61 owned by the service provider via 19. After confirming the patient information by the image reception reception tool 21, the service provider confirms by the determination tool 22 whether or not this image is a target image for determining the risk of skin diseases such as moles and melanoma. Then, when it is determined that the image is for determining the risk of skin disease, the white balance correction analyzed by the image around the skull and the scale pattern of the scale 51 sent almost at the same time as the image of the skull. The image correction tool 62 performs distortion correction from the image or the image using the scale pattern distortion information of the scale pattern image of the scale 51 sent in advance by the same user. Here, when there is no image of the color correction sheet 48 around the mole, it is also possible to perform white balance correction by using the luminance analysis of the image information of the portion other than the mole. Next, the boundary / texture extraction tool 29 that outputs information having gradation values to which the image feature analysis method using the phase stretch transform, which is the boundary detection processing step shown in Patent Document 2 and Non-Patent Document 1, is applied to the background. Exclude brightness information. Next, the image data such as the texture and morphology of the mole 12 is extracted from the image by the feature parameter extraction tool 23. After that, this image data is classified by the image classification tool 24 and the risk determination tool 25 that have been machine-learned by the case learning data, and then the determination result is notified to the user.

Since the remote skin disease risk determination system 60 shown in FIG. 16 performs processing after the mole image is in a state where the color information is not distorted due to distortion and lighting, the influence of the image distortion information on the risk determination is reduced. At the same time, it is possible to obtain the effect that the risk can be determined with the image magnification information added.

In the remote skin disease risk determination system of the present invention, an example of a process of an application performed by a user via a smartphone will be described with reference to FIG. When the user starts the application, a screen 100 requesting the image of the mole 12 is displayed as shown in FIG. 17 (a). After that, when the user attaches the member 39 that holds the relative position between the position of the smartphone and the mole to the smartphone 11 and takes a picture of the mole, the screen 101 including the image of the mole taken as shown in FIG. 17 (b) is displayed. It is displayed, and the user is requested to confirm the image such as whether the mole 12 is photographed without missing or is in focus, and the user performs the confirmation work. When the image confirmation work is completed, the image data is transferred to the service provider, and as shown in FIG. 17C, a screen 102 indicating that the image transfer is performed and the instruction to wait for the determination result is displayed is displayed. After that, as shown in FIG. 17D, the screen 103 showing the determination result is displayed. In this example, the user goes to the hospital and a screen indicating that a diagnosis at the hospital is desired is displayed.

Next, FIG. 18 shows a process of acquiring image distortion information, which is a work performed by a user using a smartphone. In the application screen 104 of the smartphone in FIG. 18, the user attaches a member 39 or a member 40 that holds the relative position between the position of the smartphone and the mole to the smartphone 11, and further attaches the scale member 51 shown in FIG. 13 to the member 39 or the member. It is an image confirmation screen after attaching to 40 and taking a picture. The user confirms that the

scales

52, 53, and 54 are captured in the focused state on this image, and transfers the captured image.

The present invention has an advantage that the burden on the patient is small because the risk of the skin disease such as melanoma can be evaluated without the patient going to the hospital. Also, on the medical institution side, there is an advantage that the burden on the medical staff is small because the equipment that has been machine-learned performs screening with learning data that reduces the risk of skin diseases.

9 ... Skin surface, 10 ... Conventional melanoma risk determination system, 11 ... Smartphone, 12 ... Mole, 19 ... Data transmission line (Internet), 20, 31, 61 ...
… Server, 21 …… Image reception reception tool, 22 …… Judgment of target image, 23 …… Feature parameter extraction tool, 24 …… Image classification tool, 25 …… Risk judgment tool, 29 …… Boundary / texture Extraction tool, 30, 60 ... Melanoma risk judgment system, 39, 40 ... Macro imaging lens, 41 ... Spacer, 41a ... Diffuse reflection processing surface, 42 ... Lens, 43 ... Clip, 44, 45 ... Clip holder, 46 ... Clip fulcrum, 47 ... Spring, 48 ... Color correction sheet, 49 color correction sheet image, 51 ... Resolution calibration scale, 51a ... Scale forming surface, 52, 53, 54 ... Scale, 62 …… Image distortion correction tool, 100, 101, 102, 103, 104 …… Application screen

Claims

It consists of a portable device having an image capturing function and a data transfer function, and a server device to which an image captured by the portable device is transferred, and the skin photographed by the portable device by a machine learning tool in the server device. In the skin disease risk judgment system that judges the disease risk from images
Before extracting the feature parameters of the image, image classification, and risk determination processing by the machine learning tool,
A remote skin disease risk determination system characterized by having a step of removing background brightness information of an image by a boundary / texture extraction tool that outputs information having a gradation value.
The remote skin disease risk determination system according to claim 1, wherein the skin disease is melanoma in the remote skin disease risk determination system.
Claim 1 or claim, wherein in the remote skin disease risk determination system, the step of removing the background brightness information of the image by the boundary / texture extraction tool is an image feature analysis method using a phase stretch transform. Item 2. The remote skin disease risk determination system according to Item 2.
In the remote skin disease risk determination system, in the skin image, the distance between the portable device and the skin for determining the disease is kept substantially constant, and a spacer member that transmits external light is attached to the portable device. The remote skin disease risk determination system according to any one of claims 1 to 3, wherein the image is a photographed image.
The remote skin disease risk determination system according to claim 4, wherein in the remote skin disease risk determination system, the spacer member is provided with an optical lens.
The remote skin disease risk determination system according to claim 4 or 5, wherein in the remote skin disease risk determination system, the spacer member has a sandy surface treated on the inside of the spacer member.
The remote skin disease risk determination system is characterized by having a step of correcting the distortion of the photographed image and acquiring the resolution information by the image information obtained by fitting the spacer member and photographing the member on which the scale is formed. The remote skin disease risk determination system according to any one of claims 4 to 6.
It consists of a portable device having an image capturing function and a data transfer function, and a server device to which an image captured by the portable device is transferred, and the skin imaged by the portable device by a machine learning tool in the server device. In the remote skin disease risk determination method that determines the disease risk from images
Before extracting the feature parameters of the image, image classification, and risk determination processing by the machine learning tool,
A remote skin disease risk determination method comprising a step of removing background brightness information of an image by a boundary / texture extraction tool that outputs information having a gradation value.
The remote skin disease risk determination method according to claim 8, wherein the skin disease is melanoma in the remote skin disease risk determination method.
Claim 8 or claim, wherein in the remote skin disease risk determination method, the step of removing the background brightness information of the image by the boundary / texture extraction tool is an image feature analysis method using a phase stretch transform. Item 9. The method for determining the risk of remote skin disease according to Item 9.
In the remote skin disease risk determination method, in the skin image, the distance between the portable device and the skin for determining the disease is kept substantially constant, and a spacer member that transmits external light is attached to the portable device. The remote skin disease risk determination method according to any one of claims 8 to 10, wherein the image is a photographed image.
The remote skin disease risk determination method according to claim 11, wherein in the remote skin disease risk determination method, the spacer member is provided with an optical lens.
The remote skin disease risk determination method according to claim 11 or 12, wherein in the remote skin disease risk determination method, the inside of the spacer member is a sandy surface to be treated.
The remote skin disease risk determination method is characterized by having a step of correcting the distortion of the photographed image and acquiring the resolution information by the image information obtained by fitting the spacer member and photographing the member on which the scale is formed. The remote skin disease risk determination method according to any one of claims 11 to 13.