A deep learning-based intelligent screening analysis method and device for tuberculin test

Abstract

The application discloses a kind of based on deep learning's tuberculin test intelligent screening analysis method and device, wherein, method includes: the image of patient's skin test is collected, and pre-processing is carried out, obtains target image, identifies the lesion area of target visible light image and marks, obtains target identification image carrying with image coding, constructs specificity identification model, identifies whether target image includes specific reaction lesion, if contains, then determine target identification image as strong positive, and carries out strong positive marking;If not contain, then according to target infrared image auxiliary hard knot identification model is identified to target identification image, judge whether to contain hard knot and / or red halo, and obtain target size;According to the relationship between target size and preset standard size, judge whether target identification image exists tuberculosis infection and infection degree.The application realizes the contactless detection of tuberculosis infection, improves the detection efficiency and detection accuracy, and helps the prevention and treatment of tuberculosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, and in particular to a tuberculosis skin test intelligent screening analysis method and device based on deep learning. BACKGROUND

[0002] Tuberculosis is a chronic infectious disease caused by Mycobacterium tuberculosis infection. Tuberculosis can invade various organs of the human body, but mainly affects the lungs, which is called pulmonary tuberculosis. The screening of tuberculosis usually adopts tuberculin test, also known as PPD (tuberculin purified protein derivative) test, which is an intradermal test for diagnosing type IV hypersensitivity reaction caused by Mycobacterium tuberculosis infection by intradermal injection of tuberculin and observing the skin condition at the injection site. Due to the high infectivity of tuberculosis, suspected cases or confirmed cases will require PPD test for people who have close contact with the patient. Therefore, tuberculosis PPD skin test has always been a commonly used auxiliary index for clinical tuberculosis diagnosis, and plays an important role in the diagnosis and differentiation of tuberculosis, and is a relatively important auxiliary examination method.

[0003] Currently, the results of PPD skin test are usually measured and diagnosed for the subjects by medical staff using traditional ruler method and finger touch method. However, the existing measurement method has the problems of not enough refined diagnosis process, long time consumption, inability to quickly screen and judge, and dependence on manual recording information, which may cause data omission or errors, etc., and is not conducive to the prevention and treatment of tuberculosis. SUMMARY

[0004] Therefore, it is necessary to provide a tuberculosis skin test intelligent screening analysis method based on deep learning in view of the above technical problems.

[0005] A tuberculosis skin test intelligent screening analysis method based on deep learning, comprising the following steps: collecting a skin test image of a patient, pre-processing the skin test image, obtaining a target image, the target image comprising a one-to-one corresponding target infrared image and target visible light image; identifying and marking a lesion area of the target visible light image to obtain a target identification image, the target identification image carrying an image code; constructing a specific identification model, identifying whether the target identification image contains a specific reaction lesion according to the specific identification model, if it contains, determining that the target identification image is strongly positive, and performing strong positive marking; when the target identification image does not contain the specific reaction lesion, identifying the target identification image according to a target infrared image auxiliary hardening identification model, judging whether the target identification image contains a hardening and / or redness, and obtaining a target size; determining whether the target identification image contains tuberculosis infection and the degree of infection according to the relationship between the target size and a preset standard size.

[0006] Furthermore, the method of collecting the patient's skin image and preprocessing the skin image to obtain the target image specifically includes: collecting the patient's skin image based on infrared light and visible light respectively, the skin image including infrared light skin image and visible light skin image, the infrared light is near-infrared light of 900nm-1000nm; normalizing the image size of the skin image; gray-scaling the normalized skin image through a specific color channel to obtain a grayscale image; binarizing the grayscale image, and performing opening and closing operations on the binarized skin image to denoise and obtain the target image.

[0007] Furthermore, the identifying and marking of the lesion area of ​​the target visible light image to obtain the target identification image specifically includes: obtaining tuberculosis image features based on historical tuberculosis images, and constructing a lesion identification model according to the tuberculosis image features; inputting the target visible light image into the lesion identification model, identifying the image features of the target visible light image through the lesion identification model, and judging whether the target visible light image contains a lesion area; if the target visible light image contains a lesion area, encoding and marking the target visible light image, determining the corresponding image code and lesion area, and obtaining the target identification image.

[0008] Furthermore, the construction of a specific recognition model and the identification of whether the target recognition image contains specific reaction lesions based on the specific recognition model specifically include: collecting image data of multiple specific reactions of blisters, necrosis and lymphangitis, the image data all containing specific reaction lesion identification areas; constructing an initial specific recognition model through a neural network, training the initial specific recognition model with the image data to obtain a specific recognition model; inputting the target recognition image into the specific recognition model to obtain a recognition result, and determining whether the target recognition image contains specific reaction lesions based on the recognition result; when the target recognition image contains specific reaction lesions, determining that the corresponding target recognition image is strongly positive and marking it as strongly positive; when the target recognition image does not contain specific reaction lesions, performing nodule recognition on the target recognition image.

[0009] Further, the target recognition image is recognized according to the target infrared image auxiliary hard node recognition model, specifically comprising: an initial hard node recognition model is constructed according to a residual network, and the initial hard node recognition model is trained and verified according to a hard node and a red halo sample set to obtain a hard node recognition model, the residual network is a resNet-V2 network based on an Inception structure; the target recognition image is input into the hard node recognition model to obtain a model recognition result; a corresponding target infrared image is obtained according to the target recognition image, a region with high absorption in the target infrared image is recognized, and an infrared recognition result is obtained; the hard node and / or red halo region is obtained according to the model recognition result auxiliary infrared recognition result; the average size of the hard node and / or red halo is detected, and the target size is obtained according to the average size of the hard node and / or red halo.

[0010] Further, the target size is obtained according to the average size of the hard node and / or red halo, specifically comprising: when the hard node or red halo exists in the target recognition image, the average size of the hard node or red halo is taken as the target size; when the hard node and red halo exist in the target recognition image, the average size of the hard node and the average size of the red halo are compared, and the larger average size is taken as the target size.

[0011] Further, the relationship between the target size and a preset standard size is used to determine whether the target recognition image has tuberculosis infection and the infection degree, specifically comprising: if the target size diameter is less than 5mm, it is determined as negative, indicating that the patient has not been infected with tuberculosis or is in the latent period of the disease; if the target size diameter is between 5-9mm, it is determined as weak positive, indicating that the patient has been infected with Mycobacterium tuberculosis or has been vaccinated with BCG; if the target size diameter is between 10-19mm, it is determined as positive, indicating that the infected person has been infected with Mycobacterium tuberculosis or has been vaccinated with BCG; if the target size diameter is greater than 20mm, it is determined as strong positive, indicating that the patient is very likely to have active pulmonary tuberculosis.

[0012] A tuberculosis skin test intelligent screening and analysis device based on deep learning, used to implement any of the above methods, comprising: a workbench, an optical component, a camera component, a fixed bracket, a housing, a computer and a display screen; the optical component, camera component and fixed bracket are all arranged in the housing; the optical component, camera component, computer and display screen are electrically connected; the workbench is used to place the patient's test part; the optical component is arranged above the workbench, including a visible light source, an infrared light source and a control power supply, the visible light source and infrared light source are arranged in an interlaced manner, and are respectively used to emit visible light and infrared light; the control power supply is used to control the turning on or off of the visible light source and the infrared light source; the fixed bracket is used to install the visible light source and the infrared light source, and a through hole is opened in the middle; the camera component is arranged above the through hole, including a lens and a camera, the camera is used to obtain visible light skin test images and infrared light skin test images according to visible light and infrared light, respectively; the computer is used to analyze whether the patient is infected with tuberculosis and the corresponding degree of infection based on the visible light skin test images and the infrared light skin test images; the display screen is used to display the screening analysis results.

[0013] Compared with the prior art, the advantages and beneficial effects of the present invention are: by collecting the patient's skin image and preprocessing the skin image, a target image is obtained, the target image includes a one-to-one corresponding target infrared image and a target visible light image, the lesion area of ​​the target visible light image is identified and marked, and a target recognition image is obtained. The target recognition image carries an image code, and a specific recognition model is constructed. According to the specific recognition model, it is identified whether the target image contains a specific reaction lesion. If it does, the target recognition image is judged to be strongly positive and marked as strongly positive; if it does not, the target recognition image is identified according to the target infrared image assisted nodule recognition model, and whether the target recognition image contains nodules and / or redness, and the target size is obtained; according to the relationship between the target size and the preset standard size, it is judged whether the target recognition image has tuberculosis infection and the degree of infection, thereby realizing contactless detection of tuberculosis infection, refining the process, shortening the detection time, and avoiding omissions or errors in manually recorded data, improving detection efficiency and detection accuracy, and contributing to the prevention and treatment of tuberculosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 1 is a flowchart of a deep learning-based intelligent screening and analysis method for tuberculosis skin tests in one embodiment;

[0015] Figure 2 1 is a target recognition image, a model recognition result of the target recognition image, and an effect diagram of the induration area of ​​the target recognition image in one embodiment.

[0016] Figure 3Schematic diagram of the structure of a tuberculosis skin test intelligent screening and analysis device based on deep learning in one embodiment;

[0017] Figure 4 for Figure 3 A schematic diagram of the structure of the workbench, camera assembly and fixed bracket;

[0018] Figure 5 for Figure 4 Schematic cross-section diagram of ;

[0019] Figure 6 This is a schematic diagram of the operating interface of a deep learning-based intelligent screening and analysis device for tuberculosis skin tests in one embodiment.

[0020] In the accompanying drawings, there are a workbench 10 , a visible light source 21 , an infrared light source 22 , a camera assembly 30 , a camera 31 , a lens 32 , a fixing bracket 40 , a housing 50 , and a display screen 60 . DETAILED DESCRIPTION

[0021] To make the present invention more clear and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0022] like Figure 1 As shown, a tuberculosis skin test intelligent screening and analysis method based on deep learning is provided, comprising the following steps:

[0023] Step S101 : collecting a skin image of a patient, pre-processing the skin image, and obtaining a target image. The target image includes a one-to-one corresponding target infrared image and a target visible light image.

[0024] Specifically, during the acquisition, the test part is imaged based on infrared light and visible light respectively through the camera equipment to obtain the skin image. After the skin image is pre-processed such as size normalization, grayscale processing, binarization processing and denoising, the target image is obtained. The target image includes a target infrared image and a target visible light image. The target infrared image corresponds to the target visible light image one-to-one, which facilitates subsequent processing of the target image and improves the recognition accuracy.

[0025] Among them, infrared light is near-infrared light of 900nm-1000nm, which can be received and imaged by ordinary CCD, and the absorption rate of water is also relatively high, which makes it convenient to identify based on the target infrared image to assist visible light imaging, thereby more accurately diagnosing the skin test results.

[0026] Step S102 : Identify and mark the lesion area of ​​the target visible light image, and obtain a target identification image. The target identification image carries an image code.

[0027] Specifically, after acquiring the target image, the lesion area in the target visible light image is identified, and the contour of the lesion area is marked to obtain a target recognition image carrying the image code, so that the lesion area within the contour can be further identified, and the patient and the target recognition image can be matched.

[0028] Step S103 , constructing a specific recognition model, and identifying whether the target recognition image contains specific reaction lesions according to the specific recognition model. If so, the target recognition image is determined to be strongly positive and marked as strongly positive.

[0029] Specifically, a specific recognition model is constructed through a neural network, and the target recognition image is input into the specific recognition model to determine whether there are specific reaction lesions in the target recognition image; if the target recognition image contains specific reaction lesions, the target recognition image is judged to be strongly positive and marked as strongly positive, prompting the patient to undergo further testing, such as CT examination, sputum smear test, etc., to determine whether the patient has active tuberculosis; if the target recognition image does not contain specific reaction lesions, step S104 is performed.

[0030] Among them, specific reaction lesions include blisters, necrosis and lymphangitis. When any of the specific reaction lesions exists, the target recognition image is determined to be strongly positive, and the patient should be prompted to undergo further testing as soon as possible to further determine whether the patient is infected with active tuberculosis.

[0031] Step S104: when the target recognition image does not contain specific reaction lesions, the target recognition image is recognized according to the target infrared image-assisted induration recognition model to determine whether the target recognition image contains induration and / or redness, and obtain the target size.

[0032] Specifically, when the target recognition image does not contain specific reaction lesions, the target recognition image is input into the nodule recognition model to obtain the model recognition result; and the target infrared image corresponding to the target recognition image is obtained, and the infrared recognition result is obtained according to the degree of water absorption in the target infrared image; the infrared recognition result is assisted by the model recognition result, and the nodules and / or red halos in the target recognition image are obtained, and the average size of the nodules and / or red halos is measured to facilitate the acquisition of the target size according to the average size of the nodules and / or red halos, so that it is possible to judge whether the patient is infected with tuberculosis and the degree of infection based on the target size and the preset standard size.

[0033] Flushes can be identified in the target recognition image based on color characteristics, while indurations typically appear darker red and can also be identified based on color characteristics. However, in some cases, indurations cannot be distinguished by the naked eye. Furthermore, since indurations are caused by fibrin deposition and the aggregation of T cells and monocytes, the complex composition of these macromolecular proteins requires complex chromatographic scanning, which cannot be simply analyzed transdermally from the image. This makes it difficult to distinguish induration areas from normal skin in the target recognition image. However, hypersensitivity reactions can cause tissue edema in the affected area, making the induration area more cellular. Furthermore, near-infrared light not only penetrates deeper into the skin, but also has a high absorption rate by water and blood. Therefore, it is possible to distinguish indurations, flushes, and surrounding normal skin based on the degree of water absorption in the target infrared image. If the water absorption level in the induration area is higher, the corresponding area can be marked in the target infrared image, representing the initial induration area. Combined with the model recognition results, the actual induration area can be further determined, improving recognition accuracy.

[0034] Step S105 , judging whether the target recognition image has tuberculosis infection and the degree of infection based on the relationship between the target size and the preset standard size.

[0035] Specifically, after obtaining the target size, the relationship between the target size and the corresponding preset standard size is used to determine whether the target recognition image has tuberculosis infection and the degree of infection, thereby achieving rapid and accurate diagnosis of tuberculosis infection, realizing contactless detection of tuberculosis infection, and improving detection efficiency.

[0036] In this embodiment, a target image is acquired by collecting a patient's skin image and preprocessing the skin image. The target image includes a one-to-one corresponding target infrared image and a target visible light image. The lesion area of ​​the target visible light image is identified and marked to acquire a target recognition image. The target recognition image carries an image code, and a specific recognition model is constructed. The target image is identified according to the specific recognition model to determine whether it contains a specific reaction lesion. If so, the target recognition image is determined to be strongly positive and marked as strongly positive. If not, the target recognition image is identified according to the target infrared image-assisted nodule recognition model to determine whether the target recognition image contains nodules and / or redness, and the target size is obtained. Based on the relationship between the target size and the preset standard size, it is determined whether the target recognition image has tuberculosis infection and the degree of infection, thereby realizing contactless detection of tuberculosis infection, refining the process, shortening the detection time, and avoiding omissions or errors in manually recorded data, improving detection efficiency and detection accuracy, and contributing to the prevention and treatment of tuberculosis.

[0037] Among them, step S101 specifically includes: normalizing the image size of the skin image; gray-scaling the normalized skin image through a specific color channel to obtain a grayscale image; binarizing the grayscale image, and performing opening and closing operations on the binarized skin image to denoise and obtain a target image.

[0038] Specifically, after collecting and acquiring the skin image, the size of the skin image is normalized to ensure size consistency for easy subsequent processing, and the size-normalized image is grayscaled through a specific color channel to obtain a grayscale image; the grayscale image is binarized, and the binarized skin image is opened and closed for denoising to obtain a target image, thereby enabling more accurate identification and processing of the target image, thereby improving the accuracy of tuberculosis screening analysis.

[0039] Among them, the opening and closing operations are both based on the combination of dilation and erosion operations. The opening operation adopts the method of first erosion and then dilation, which is used to eliminate small objects, separate objects at thin points, and smooth the boundaries of larger objects without significantly changing their area to extract horizontal or vertical lines; the closing operation adopts the method of first dilation and then erosion, which is used to fill small holes in objects, connect adjacent objects, and smooth their boundaries without significantly changing their area.

[0040] Among them, step S102 specifically includes: obtaining tuberculosis image features based on historical tuberculosis images, and constructing a lesion recognition model based on the tuberculosis image features; inputting the target visible light image into the lesion recognition model, and identifying the image features of the target visible light image through the lesion recognition model to determine whether the target visible light image contains a lesion area; if the target visible light image contains a lesion area, encoding and marking the target visible light image, determining the corresponding image code and lesion area, and obtaining a target recognition image.

[0041] Specifically, a plurality of historical tuberculosis images are obtained as samples, tuberculosis image features in the historical tuberculosis images are extracted, a lesion recognition model is constructed according to the tuberculosis image features, the target visible light image is input into the lesion recognition model, and the image features in the target visible light image are recognized by the lesion recognition model to determine whether there is a corresponding lesion area in the target visible light; if there is a lesion area, the lesion area is outlined, and a target recognition image with the outline mark is obtained, and the target recognition image is image-encoded at the same time, so as to associate the patient with the target recognition image; if there is no lesion area in the target recognition image, the recognition is completed, and whether the patient is infected with tuberculosis is determined based on whether there is a specific reaction in the target recognition image.

[0042] Among them, step S103 specifically includes: collecting image data of specific reactions of multiple blisters, necrosis and lymphangitis, where the image data all contain specific reaction lesion identification areas; constructing an initial specific recognition model through a neural network, training the initial specific recognition model through image data, and obtaining a specific recognition model; inputting a target recognition image into the specific recognition model to obtain a recognition result, and determining whether the target recognition image contains specific reaction lesions based on the recognition result; when the target recognition image contains specific reaction lesions, determining that the corresponding target recognition image is strongly positive and marking it as strongly positive; when the target recognition image does not contain specific reaction lesions, performing nodule recognition on the target recognition image.

[0043] Specifically, image data of multiple specific reactions such as blisters, necrosis and lymphangitis are collected, and the image data all contain specific reaction lesion identification areas; an initial specific recognition model is constructed through a neural network, and the initial specific recognition model is trained and verified using image data to obtain a specific recognition model; the target recognition image is input into the specific recognition model to obtain a corresponding recognition result, which includes the presence or absence of a specific reaction lesion; when a specific reaction lesion exists in the target recognition image, the corresponding target recognition image is determined to be strongly positive, prompting the patient to undergo further tuberculosis testing as soon as possible, such as a CT scan, sputum smear examination, etc., so as to accurately determine whether the patient has active tuberculosis; when no specific reaction lesions exist in the target recognition image, the target recognition image is identified for nodules and red halos, and whether the patient is infected with tuberculosis and the corresponding degree of infection are determined based on the identification of nodules and red halos, thereby achieving tuberculosis screening for patients.

[0044] Among them, step S104 specifically includes: constructing an initial nodule recognition model based on the residual network, and training and verifying the initial nodule recognition model based on the sample set of nodules and blushes to obtain a nodule recognition model, and the residual network is a resNet-V2 network based on the Inception structure; inputting the target recognition image into the nodule recognition model to obtain the model recognition result; obtaining the corresponding target infrared image according to the target recognition image, identifying the area with high absorption in the target infrared image, and obtaining the infrared recognition result; assisting the infrared recognition result according to the model recognition result to obtain the nodule and / or blush area; detecting the average size of the nodules and / or blushes, and obtaining the target size according to the average size of the nodules and / or blushes.

[0045] Specifically, when the induration cannot be identified based on color features, an initial induration recognition model is constructed based on the residual network of the resNet-V2 network with the Inception structure. Samples of induration and blush are used to train and verify the initial induration recognition model. The induration recognition model can identify the blush area by color features.Figure 2 The target recognition image shown in the first column of images is input into a hard nodule recognition model to obtain a model recognition result, as shown in the second column of images. Figure 2 The target recognition image shown in the first column of images is input into a hard nodule recognition model to obtain a model recognition result, as shown in the second column of images. Figure 2 The target recognition image shown in the first column of images is input into a hard nodule recognition model to obtain a model recognition result, as shown in the second column of images.

[0046] In the case where the target recognition image contains both a hard nodule and a red halo, the average size of the hard nodule and the average size of the red halo are compared, and the larger one is taken as the target size.

[0047] Specifically, since the target recognition image can contain a red halo, a hard nodule, or both a red halo and a hard nodule, in the case where only one of a red halo and a hard nodule exists, the average size of the red halo or the hard nodule is taken as the target size; in the case where both a red halo and a hard nodule exist, the average size of the red halo and the average size of the hard nodule are compared, and the larger one is taken as the target size, so that the target size can be used to accurately analyze and determine the tuberculosis infection, and the detection efficiency is improved.

[0048] If the target size diameter is less than 5 mm, it is determined to be negative, indicating that the patient has not been infected with tuberculosis or is in the latent period of the disease; if the target size diameter is between 5 mm and 9 mm, it is determined to be weakly positive, indicating that the patient has been infected with Mycobacterium tuberculosis or has been vaccinated with BCG; if the target size diameter is between 10 mm and 19 mm, it is determined to be positive, indicating that the infected person has been infected with Mycobacterium tuberculosis or has been vaccinated with BCG; and if the target size diameter is greater than 20 mm, it is determined to be strongly positive, indicating that the patient is very likely to have active pulmonary tuberculosis.

[0049] Specifically, when the target size is less than 5mm, the target recognition image is determined to be negative, indicating that the patient is not infected with tuberculosis or is in the incubation period of the disease; if the target size diameter is between 5-9mm, the target recognition image is determined to be weakly positive, indicating that the patient has been infected with Mycobacterium tuberculosis or vaccinated with BCG; if the target size diameter is between 10-19mm, the target recognition image is determined to be positive, indicating that the infected person has been infected with Mycobacterium tuberculosis or vaccinated with BCG, but does not mean that the patient has tuberculosis, and further testing is needed to determine whether the infection is present; if the target size diameter is greater than 20mm, it is determined to be strongly positive, indicating that the patient is very likely to have active pulmonary tuberculosis and needs to undergo further testing as soon as possible to determine the infection result and degree of infection.

[0050] In one embodiment, Figure 3 and Figure 5 As shown, the present invention also provides a tuberculosis skin test intelligent screening and analysis device based on deep learning, which is used to implement the above-mentioned tuberculosis skin test intelligent screening and analysis method based on deep learning, including: a workbench 10, an optical component, a camera component 30, a fixed bracket 40, a shell 50, a computer (not shown) and a display screen 60; the optical component, the camera component 30 and the fixed bracket 40 are all arranged in the shell 50; the optical component, the camera component 30, the computer and the display screen 60 are electrically connected; the workbench 10 is used to place the patient's test part; the optical component is arranged above the workbench 10, and includes a visible light source 21, an infrared light source 22 and a control circuit The visible light source 21 and the infrared light source 22 are arranged in an interlaced manner, and are respectively used to emit visible light and infrared light; the control power supply is used to control the opening or closing of the visible light source 21 and the infrared light source 22; the fixing bracket 40 is used to install the visible light source 21 and the infrared light source 22, and a through hole is opened in the middle; the camera component 30 is arranged above the through hole, including a camera 31 and a lens 32, and the camera 31 is used to obtain visible light skin images and infrared light skin images according to visible light and infrared light respectively; the computer is used to analyze whether the patient is infected with tuberculosis and the corresponding degree of infection based on the visible light skin images and the infrared light skin images; the display screen 60 is used to display the screening analysis results.

[0051] In this embodiment, the patient places the test part on the workbench 10, and the visible light source 21 and the infrared light source 22 are turned on respectively. The camera 31 is used to collect the visible light skin image and the infrared skin image of the patient respectively, and the above method is implemented by a computer to process the visible light skin image and the infrared skin image. The presence of tuberculosis infection and the corresponding degree of infection of the patient are analyzed according to the processing results. Finally, the analysis results are displayed on the display screen 60, thereby realizing contactless tuberculosis infection screening and improving screening efficiency and diagnostic accuracy.

[0052] In one embodiment, the process of a patient undergoing testing using the aforementioned deep learning-based intelligent tuberculosis skin test screening and analysis device is as follows:

[0053] First, the patient places the test area flat in front of the camera. The interface displays a real-time image of the arm and, under the guidance of medical staff, places it in the correct position on the workbench.

[0054] Secondly, after the doctor confirms the diagnosis, the camera is controlled to capture multiple images, including infrared skin images and visible light images, and the clearest image is intelligently selected for algorithm processing;

[0055] Again, the original image, processed image and diagnostic information are taken, such as Figure 2 As shown, it is transferred to the operation interface and all original data is saved to the local hard disk.

[0056] Finally, the doctor confirms and gives the final diagnosis, which is manually entered into the patient's electronic medical record. The doctor can choose whether to print out the diagnosis for the patient.

[0057] In one embodiment, the interface operation of the above-mentioned tuberculosis skin test intelligent screening and analysis device based on deep learning is as follows: Figure 6 As shown, the corresponding operation steps are as follows:

[0058] Click "Enter" to switch to real-time image display (control the white light to see the light turn on);

[0059] Click "Start" to capture and save the image. Two images are displayed on the screen: the left one is the original image, and the right one is the image processed by the algorithm. The detection results are given at the top of the screen (when the shooting results are displayed, the white light is turned off and the infrared light is on, and the infrared image is captured and saved);

[0060] If the doctor is not satisfied with the test data, click "Measure" and use the mouse to draw a straight line in the image. The length value will be displayed and the examination result at the top of the screen will be changed according to the current length value.

[0061] Click "Finish" to switch to real-time image display (control the infrared light to turn off and the white light to turn on);

[0062] Click "Exit" to close the software window (control the light source to turn off).

[0063] Through the above-mentioned detection process and operation method, refined contactless screening for tuberculosis infection can be achieved, shortening the diagnosis time for both doctors and patients, improving diagnosis efficiency, and contributing to the prevention and treatment of tuberculosis; and high-definition images can be collected in real time through the terminal, and the collected images can be quickly and accurately analyzed by computers using multiple algorithms to improve the accuracy of screening; in addition, doctors can also remotely control through the Internet of Things terminal, facilitating data input and management, and avoiding data errors.

[0064] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0065] Obviously, those skilled in the art should understand that the modules or steps of the present invention described above can be implemented using a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices. Alternatively, they can be implemented using program codes executable by the computing device, so that they can be stored in a computer storage medium (ROM / RAM, magnetic disk, optical disk) and executed by the computing device. In some cases, the steps shown or described can be performed in a different order than herein, or they can be made into individual integrated circuit modules, or multiple modules or steps can be made into a single integrated circuit module for implementation. Therefore, the present invention is not limited to any specific combination of hardware and software.

[0066] The above is only an embodiment of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the field can improve and implement this scheme in combination with their own abilities under the inspiration given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A tuberculosis skin test intelligent screening and analysis method based on deep learning, characterized in that: The following steps are involved: Acquiring a skin image of the patient, preprocessing the skin image, and acquiring a target image, wherein the target image includes a one-to-one corresponding target infrared image and a target visible light image; Identifying and marking the lesion area of ​​the target visible light image to obtain a target recognition image, wherein the target recognition image carries an image code; Constructing a specific recognition model, and identifying whether the target recognition image contains a specific reaction lesion according to the specific recognition model; if so, determining that the target recognition image is strongly positive and marking it as strongly positive; When the target recognition image does not contain a specific reaction lesion, the target recognition image is recognized according to the target infrared image-assisted induration recognition model to determine whether the target recognition image contains induration and / or redness, and to obtain the target size, including: inputting the target recognition image into the induration recognition model to obtain a model recognition result, and obtaining a corresponding target infrared image; obtaining an infrared recognition result according to the degree of water absorption in the target infrared image; assisting the infrared recognition result according to the model recognition result to obtain the induration and / or redness in the target recognition image and the target size thereof; wherein, the degree of water absorption in the induration area is higher, marking the corresponding area in the target infrared image, obtaining a preliminarily determined induration area, and determining the actual induration area in combination with the model recognition result; Based on the relationship between the target size and the preset standard size, it is determined whether the target recognition image has tuberculosis infection and the degree of infection.

2. The tuberculosis skin test intelligent screening and analysis method based on deep learning according to claim 1, characterized in that: The collecting of the patient's skin image, preprocessing of the skin image, and obtaining of the target image specifically include: Collecting skin images of the patient based on infrared light and visible light respectively, wherein the skin images include infrared skin images and visible light skin images, and the infrared light is near-infrared light of 900nm-1000nm; performing image size normalization processing on the skin image; Grayscale the normalized skin image through a specific color channel to obtain a grayscale image; The grayscale image is binarized, and the binarized skin image is subjected to an opening and closing operation to remove noise, so as to obtain a target image.

3. The tuberculosis skin test intelligent screening and analysis method based on deep learning according to claim 1, characterized in that: The step of identifying and marking the lesion area in the target visible light image to obtain a target identification image specifically includes: Acquiring tuberculosis image features based on historical tuberculosis images, and building a lesion recognition model based on the tuberculosis image features; Inputting the target visible light image into a lesion recognition model, identifying image features of the target visible light image through the lesion recognition model, and determining whether the target visible light image contains a lesion area; If the target visible light image contains a lesion area, the target visible light image is encoded and marked, the corresponding image code and lesion area are determined, and a target recognition image is obtained.

4. The method for intelligent screening and analysis of tuberculosis skin test based on deep learning according to claim 1, characterized in that: The constructing of the specific recognition model and identifying whether the target recognition image contains a specific reaction lesion according to the specific recognition model specifically includes: Collecting a plurality of image data of specific reactions of blisters, necrosis and lymphangitis, wherein the image data all contain specific reaction lesion identification areas; constructing an initial specific recognition model through a neural network, and training the initial specific recognition model through the image data to obtain a specific recognition model; Inputting the target recognition image into the specific recognition model to obtain a recognition result, and determining whether the target recognition image contains a specific reaction lesion according to the recognition result; When the target recognition image contains a specific reaction focus, the corresponding target recognition image is determined to be strongly positive and marked as strongly positive; When the target recognition image does not include a specific reaction focus, induration recognition is performed on the target recognition image.

5. The tuberculosis skin test intelligent screening and analysis method based on deep learning according to claim 1, characterized in that: The identifying of the target recognition image according to the target infrared image-assisted induration recognition model specifically includes: An initial induration recognition model is constructed based on a residual network, and the initial induration recognition model is trained and verified based on a sample set of indurations and redness to obtain an induration recognition model, wherein the residual network is a resNet-V2 network based on an Inception structure; Inputting the target recognition image into a nodule recognition model to obtain a model recognition result; Acquire a corresponding target infrared image according to the target recognition image, identify a region with high absorption in the target infrared image, and obtain an infrared recognition result; Acquire the induration and / or redness area according to the infrared recognition result assisted by the model recognition result; The average size of the induration and / or redness is detected, and the target size is obtained according to the average size of the induration and / or redness.

6. The method for intelligent screening and analysis of tuberculosis skin test based on deep learning according to claim 5, characterized in that: The obtaining of a target size according to the average size of the induration and / or redness specifically includes: When there are nodules or red halos in the target recognition image, the average size of the nodules or red halos is used as the target size; When both induration and redness exist in the target recognition image, the average size of the induration and the average size of the redness are compared, and the larger average size is taken as the target size.

7. The method for intelligent screening and analysis of tuberculosis skin test based on deep learning according to claim 1, characterized in that: The determining, based on the relationship between the target size and the preset standard size, whether the target recognition image has tuberculosis infection and the degree of infection specifically includes: If the target size is less than 5 mm in diameter, it is considered negative, indicating that the patient is not infected with tuberculosis or is in the latent stage of the disease; If the target size diameter is between 5-9 mm, it is considered weakly positive, indicating that the patient has been infected with Mycobacterium tuberculosis or vaccinated with BCG; If the target size diameter is between 10-19 mm, it is considered positive, indicating that the infected person has been infected with Mycobacterium tuberculosis or vaccinated with BCG; If the target size is larger than 20 mm in diameter, it is considered a strong positive, indicating that the patient is very likely to have active pulmonary tuberculosis.

8. A tuberculosis skin test intelligent screening and analysis device based on deep learning, used to implement the method of any one of claims 1 to 7, characterized in that: include: A workbench, an optical component, a camera component, a fixed bracket, a housing, a computer and a display screen; the optical component, the camera component and the fixed bracket are all arranged in the housing; the optical component, the camera component, the computer and the display screen are electrically connected; the workbench is used to place the patient's test part; the optical component is arranged above the workbench, and includes a visible light source, an infrared light source and a control power supply, the visible light source and the infrared light source are arranged in an interlaced manner, and are respectively used to emit visible light and infrared light; the control power supply is used to control the opening or closing of the visible light source and the infrared light source; the fixed bracket is used to install the visible light source and the infrared light source, and a through hole is opened in the middle; the camera component is arranged above the through hole, and includes a lens and a camera, and the camera is used to obtain visible light skin images and infrared light skin images according to visible light and infrared light respectively; the computer is used to analyze whether the patient is infected with tuberculosis and the corresponding degree of infection based on the visible light skin images and the infrared light skin images; the display screen is used to display the screening analysis results.

Images