Wound surface bacterial autofluorescence detection system and method

Abstract

The present invention discloses a wound surface bacteria autofluorescence detection system and method, wherein an excitation light source module is used to irradiate the wound surface with a laser light source of a specific wavelength to excite the autofluorescence of the wound surface bacteria; an image acquisition module is used to acquire the autofluorescence image of the wound surface bacteria, and send the acquired fluorescence image to a host computer; the host computer is used to enhance the acquired fluorescence image based on an improved Retinex‑Net image enhancement processing algorithm with residual connection, so as to perform bacteria detection based on the obtained fluorescence enhanced image. The present invention greatly improves the quality of fluorescence images based on the improved Retinex‑Net algorithm, and can obtain clear bacterial fluorescence images in low-light environments, improve detection speed and accuracy, and can provide bacteria distribution information in real time to assist clinical decision-making; the system is compact in design, small in size, easy to carry and use, and suitable for clinical field detection.

Description

Technical Field

[0001] The present invention relates to the technical field of wound surface bacteria detection, and in particular to a wound surface bacteria autofluorescence detection system and method. Background Art

[0002] As the most common complication in wound care, bacterial infection is widely present in the treatment of burns or traumatic wounds, and often leads to longer hospitalization time and higher mortality. Immediate identification and early treatment of infection are the key to achieving wound healing. There are currently two main methods for diagnosing bacterial infection: clinicians evaluate patients' clinical symptoms and signs (CSS) and microbial swab sampling and culture. The evaluation of infection based on common clinical symptoms and signs of infection is often subjective, and doctors with different experience may come to different conclusions, lacking sufficient accuracy. The standard for infection diagnosis is to determine the bacterial load and type of the wound by surface swab sampling and culture, but the selection of sampling areas and different sampling methods may miss certain bacterial load areas, resulting in false negative results. In addition, the swab sampling method often takes a long time to obtain test results, at which time the infection may have spread, or the dominant microbial community may have changed, resulting in large errors in the test results. In summary, a new technology is urgently needed in clinical bacterial detection and infection judgment to assist clinicians in quickly identifying the presence and load of bacteria.

[0003] Studies have shown that living bacteria have a large number of intracellular biomolecules related to energy-producing reactions. These endogenous molecules can spontaneously produce specific fluorescence when excited by specific wavelengths. This fluorescence produced by endogenous biological fluorophores is an inherent property of bacteria and is called bacterial autofluorescence. The bacterial autofluorescence imaging method is a new bacterial detection method. Based on the characteristic of bacterial autofluorescence, combined with fluorescence imaging processing technology, it can realize the visualization of wound bacteria detection, and then realize real-time detection of bacterial load, detection of bacterial biofilm, guidance of debridement, and help wound care points to achieve antimicrobial drug management and other functions. In addition, fluorescence imaging technology can also accelerate the healing of chronic wounds and effectively save the treatment costs of wound care points. However, the existing bacterial autofluorescence imaging technology has the following defects:

[0004] (1) Insufficient sensitivity and resolution: Existing fluorescence imaging devices have poor imaging quality in low-light environments, making it difficult to accurately identify bacteria in actual clinical settings.

[0005] (2) Complex equipment and high cost: Existing fluorescence imaging systems are usually large in size, complex to operate, and have high equipment costs, which is not conducive to widespread promotion and application.

[0006] (3) Low data processing efficiency: Traditional image processing algorithms have limited effectiveness in processing fluorescence images, resulting in insufficient accuracy and real-time performance of the detection results. Summary of the invention

[0007] The present application provides a wound bacterial autofluorescence detection system and method to solve the problems of the existing bacterial autofluorescence detection technology, such as insufficient sensitivity and resolution, complex equipment and high cost, and low data processing efficiency.

[0008] According to the first aspect, an embodiment provides a wound bacterial autofluorescence detection system, the system comprising an excitation light source module, an image acquisition module and a host computer;

[0009] The excitation light source module is used to use a laser light source of a specific wavelength to irradiate the wound surface to stimulate the autofluorescence of bacteria on the wound surface;

[0010] The image acquisition module is used to acquire the autofluorescence image of the bacteria on the wound surface and send the acquired fluorescence image to the host computer;

[0011] The host computer is used to enhance the collected fluorescence image based on the improved Retinex-Net image enhancement processing algorithm with residual connection, so as to perform bacteria detection based on the obtained fluorescence enhanced image.

[0012] Furthermore, the excitation light source module includes an excitation light source and a light source driving module connected to the excitation light source;

[0013] The excitation light source uses LED lamp beads with a peak wavelength of a preset value;

[0014] The irradiation radius of a single lamp bead is:

[0015] r = tanθ × d

[0016] Where θ is the half angle of the light source divergence angle, d is the distance from the illuminated surface to the light source, and the illuminated area of ​​the illuminated surface is expressed as:

[0017] S=π×(tanθ×d)2

[0018] Where S is the area of ​​the surface irradiated by the light source. When the radiation flux of the light source is known, The radiation flux or irradiance of the incident surface is expressed as:

[0019]

[0020] The light source driving module includes a power chip and a constant current driver, and changes the brightness of the light source by controlling the output power.

[0021] Further, the image acquisition module includes a camera and a dual-band bandpass filter;

[0022] The camera is used to collect autofluorescence images of bacteria on the wound surface. When shooting, the camera is 10-30 cm away from the wound surface, and the lens is a zoom lens with a focal length of 8-50 mm;

[0023] The dual-band bandpass filter is used to filter out light in bands other than bacterial autofluorescence during image acquisition.

[0024] Furthermore, the acquired fluorescence image is enhanced based on the improved Retinex-Net image enhancement processing algorithm with residual connection, specifically including:

[0025] Image enhancement processing is performed based on a pre-trained convolutional neural network, wherein the convolutional neural network includes a Decom-Net subnetwork and an Enhance-Net subnetwork. The image enhancement processing steps include:

[0026] The input fluorescence image I is decomposed into a reflectance image and an illumination image through the Decom-Net sub-network:

[0027]

[0028] in, and are the estimated reflectance image and illumination image, respectively;

[0029] Shared weights: The network weights are shared between the reflectance image and the illumination image to maintain consistency;

[0030] The brightness of the illuminated image is adjusted and enhanced through the Enhance-Net sub-network based on the codec architecture:

[0031]

[0032] Residual connection: Residual connection is introduced into the network to solve the gradient disappearance problem and improve the efficiency and stability of network training. The residual connection formula is as follows:

[0033]

[0034] Denoising: Denoising is performed on the reflectivity image to eliminate noise in the image and improve image clarity and detail retention:

[0035]

[0036] Image reconstruction: The brightness-adjusted illumination image and the denoised reflectance image are recombined to generate the final enhanced fluorescence image:

[0037]

[0038] Combination strategy: Through the weighted combination strategy, the features of the illumination image and the reflectance image are fused to achieve global brightness adjustment and detail enhancement.

[0039] According to the second aspect, an embodiment provides a method for detecting wound bacterial autofluorescence, the method comprising:

[0040] A laser light source with a specific wavelength is used to illuminate the wound surface to stimulate the autofluorescence of bacteria on the wound surface;

[0041] Collect autofluorescence images of bacteria on the wound surface;

[0042] The collected fluorescence images are enhanced by using an improved Retinex-Net image enhancement processing algorithm based on residual connections, so that bacteria can be detected based on the obtained fluorescence enhanced images.

[0043] Furthermore, collecting the autofluorescence image of the bacteria on the wound surface specifically includes:

[0044] During image acquisition, a dual-band bandpass filter was used to filter out light other than bacterial autofluorescence.

[0045] Furthermore, the collected fluorescence image is enhanced by using an improved Retinex-Net image enhancement processing algorithm based on residual connections, specifically including:

[0046] Image enhancement processing is performed based on a pre-trained convolutional neural network, wherein the convolutional neural network includes a Decom-Net subnetwork and an Enhance-Net subnetwork. The image enhancement processing steps include:

[0047] The input fluorescence image I is decomposed into a reflectance image and an illumination image through the Decom-Net sub-network:

[0048]

[0049] in, and are the estimated reflectance image and illumination image, respectively;

[0050] Shared weights: The network weights are shared between the reflectance image and the illumination image to maintain consistency;

[0051] The brightness of the illuminated image is adjusted and enhanced through the Enhance-Net sub-network based on the codec architecture:

[0052]

[0053] Residual connection: Residual connection is introduced into the network to solve the gradient disappearance problem and improve the efficiency and stability of network training. The residual connection formula is as follows:

[0054]

[0055] Denoising: Denoising is performed on the reflectivity image to eliminate noise in the image and improve image clarity and detail retention:

[0056]

[0057] Image reconstruction: The brightness-adjusted illumination image and the denoised reflectance image are recombined to generate the final enhanced fluorescence image:

[0058]

[0059] Combination strategy: Through the weighted combination strategy, the features of the illumination image and the reflectance image are fused to achieve global brightness adjustment and detail enhancement.

[0060] Furthermore, before enhancing the collected fluorescence image based on the improved Retinex-Net image enhancement processing algorithm with residual connection, the method further includes:

[0061] Constructing a wound surface bacterial fluorescence image sample dataset, and dividing the sample dataset into a training set and a test set;

[0062] The training set and the test set are used to train and test a plurality of pre-selected image enhancement processing algorithm models, and a plurality of evaluation indicators are used to evaluate the trained plurality of image enhancement processing algorithm models, and the best image enhancement processing algorithm model is obtained according to the evaluation results.

[0063] Furthermore, the pre-selected multiple image enhancement processing algorithm models include SSR, MSR, MSRCR algorithms, Retinex-Net algorithm and improved Retinex-Net algorithm with residual connection;

[0064] The evaluation indicators include structural similarity measurement SSIM, peak signal-to-noise ratio PSNR, information entropy EN, and average gradient AG.

[0065] Furthermore, a wound bacterial fluorescence image sample dataset is constructed, specifically including:

[0066] 1 ml of bacterial suspension with a concentration of 109 cfu / ml was mixed with 9 ml of saline solution in a 10 ml glass beaker to dilute and make a bacterial solution with a concentration of 106 cfu / ml.

[0067] Place the glass beaker in a vortex oscillator and oscillate for 2 minutes to evenly distribute the bacteria and ensure that sufficient sampling is done each time. During the stirring process, keep the solution at 10 ml.

[0068] Use a medical plastic pipette to draw 1 ml each time and drop it onto the biological tissue phantom, turn on the excitation light source and adjust the angle and distance to illuminate the contaminated area to observe the fluorescence phenomenon;

[0069] Open the host computer software, configure the image acquisition module, and conduct a blank experiment to eliminate the interference of fluorescence imaging of the biological tissue phantom itself;

[0070] Fluorescence images of biological tissue phantoms under different bacterial concentrations were collected to construct a sample data set.

[0071] The present application provides a wound surface bacterial autofluorescence detection system and method, the system includes an excitation light source module, an image acquisition module and a host computer; the excitation light source module is used to use a laser light source of a specific wavelength to irradiate the wound surface to excite the autofluorescence of the wound surface bacteria; the image acquisition module is used to collect the autofluorescence image of the wound surface bacteria and send the collected fluorescence image to the host computer; the host computer is used to enhance the collected fluorescence image based on the improved Retinex-Net image enhancement processing algorithm with residual connection, so as to perform bacteria detection based on the obtained fluorescence enhanced image. It has the following beneficial effects:

[0072] (1) High sensitivity and high resolution: The improved Retinex-Net algorithm greatly improves the quality of fluorescence images, and can obtain clear bacterial fluorescence images even in low-light environments.

[0073] (2) Portability and ease of use: The system is compact in design and small in size, making it easy to carry and use, and suitable for clinical field testing.

[0074] (3) Efficient data processing: The use of advanced image processing algorithms improves detection speed and accuracy, and can provide real-time bacterial distribution information to assist clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 A structural diagram of a wound surface bacterial autofluorescence detection system provided by one embodiment of the present invention;

[0076] Figure 2 A PCB diagram of a light source lamp board in a wound surface bacteria autofluorescence detection system provided by one embodiment of the present invention;

[0077] Figure 3 A structural diagram of a Retinex-Net algorithm with residual connections in a wound bacterial autofluorescence detection system provided by an embodiment of the present invention;

[0078] Figure 4 The results of a bacterial fluorescence experiment on muscle parts in a wound bacterial autofluorescence detection system provided by an embodiment of the present invention;

[0079] Figure 5 The results of a bacterial fluorescence experiment at the tendon site in a wound bacterial autofluorescence detection system provided by an embodiment of the present invention;

[0080] Figure 6 A flow chart of a method for detecting bacterial autofluorescence on a wound surface provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0081] The present invention is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are for making the present application better understood. However, those skilled in the art can easily recognize that some features can be omitted in different situations, or can be replaced by other elements, materials, methods. In some cases, some operations related to the present application are not shown or described in the specification, this is to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations according to the description in the specification and the general technical knowledge in the art.

[0082] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0083] The first embodiment of the present invention provides a wound surface bacterial autofluorescence detection system, such as Figure 1 As shown, the system includes an excitation light source module, an image acquisition module and a host computer.

[0084] The excitation light source module is used to illuminate the wound surface with a laser light source of a specific wavelength to stimulate the autofluorescence of bacteria on the wound surface; the image acquisition module is used to acquire the autofluorescence image of bacteria on the wound surface and send the acquired fluorescence image to the host computer; the host computer is used to enhance the acquired fluorescence image based on the improved Retinex-Net image enhancement processing algorithm with residual connection, so as to perform bacteria detection based on the obtained fluorescence enhanced image. The specific contents are as follows:

[0085] (1) Excitation light source module design

[0086] Excitation light source type. There are two main types of current mainstream fluorescence excitation light sources: Laser Diode (LD) light source and Light Emitting Diode (LED) light source. However, the LD light source has a high luminous intensity, especially the single-point intensity is very high, which may be dangerous to fragile wound tissues. In addition, the LD light source generates a lot of heat, which places extremely high demands on the heat dissipation function of the system. Taking all factors into consideration, the LED light source is more suitable as the excitation light source for this system.

[0087] LED light source lamp board design. The light source uses LED lamp beads with a peak wavelength of 405nm, a single power of 1w, a radiation flux of <200mw, and a light divergence half angle of 10° for a single lamp bead.

[0088] The irradiation radius of a single lamp bead is:

[0089] r = tanθ × d

[0090] Where θ is the half angle of the light source divergence angle, d is the distance from the illuminated surface to the light source, and the illuminated area of ​​the illuminated surface is expressed as:

[0091] S=π×(tanθ×d)2

[0092] Where S is the area of ​​the surface irradiated by the light source. When the radiation flux of the light source is known, The radiation flux or irradiance of the incident surface is expressed as:

[0093]

[0094] When θ is 10° and d is 10cm, When the power is 500mw, the formula shows that the intensity of the excitation light on the surface of the irradiated object is about 51mw / cm2, the irradiation diameter is 3.5cm, and the irradiation area is 9.8cm2, which can cover the general wound area.

[0095] like Figure 2 The figure shows the PCB diagram of the light board drawn by Altium designer software. The hollow diameter of the half-moon PCB light board is the outer diameter of the camera lens, which can just pass through the camera lens, reducing the system volume while concentrating the excitation light source on the center of the camera's field of view.

[0096] The light source driving circuit is mainly composed of the SB6286 power chip and the BP1360LED constant current driver. The brightness of the light source is changed by controlling the output power. The working state of the SB6286 is controlled by the pin EN. The chip works when the input is high, and the switch is disconnected when the input is low. Connect EN to the input power supply for automatic startup. The BP1360 input voltage range is from 5V to 30V, and the output current is set by the sampling resistor. The maximum output current of a single LED can reach 1000 mA. The BP1360 accepts 0.5-2.5V analog dimming and PWM dimming with a wide frequency range through the DIM pin, and can drive multiple 2W LEDs.

[0097] (2) Image acquisition module design

[0098] Camera selection. When collecting fluorescent images of bacteria on the wound surface, it is necessary to select a camera that has high sensitivity to specific wavelengths of light (such as 500-545nm and 600-665nm).

[0099] Sony's IMX585 camera has many advantages: first, the high resolution it provides can capture the fine details in the image; second, the high sensitivity allows the camera to effectively capture weak fluorescence signals in low-light environments, reducing exposure time while maintaining image quality; in addition, the low-noise design improves the signal-to-noise ratio of the image and ensures image clarity; the fast readout speed helps capture clear images in dynamic processes. Finally, the IMX585 has good compatibility, which greatly simplifies the integration and use process.

[0100] Table 1 Camera parameters

[0101]

[0102] Lens selection. Because the system uses an integrated design, the fluorescence imaging camera is used at the same distance as the light source, 10-30cm from the wound. In order to obtain a larger (≧60×60mm) field of view, there are special requirements for the focal length of the camera lens. The relationship between the field of view and the focal length is as follows:

[0103]

[0104] When the object distance is 10-30cm, the camera CCD size is 7mm*5.3mm, and the field of view shooting area is 64mm*48mm, the focal length range is calculated to be 10.9-32.8, so the lens focal length finally selected by this system is a zoom lens with a focal length of 8-50mm.

[0105] Filter selection. The bacteria commonly found on the wound surface all emit red fluorescence (in the range of 600-665nm) under the irradiation of a 405nm laser. Only a few species such as Pseudomonas aeruginosa produce cyan fluorescence (500-545nm). Therefore, this device uses a dual-band bandpass filter.

[0106] (3) Host computer software design

[0107] In order to realize the function of real-time detection and processing, the program control interface is developed based on the python language, camera SDK and openCV, cuda, and PyQT. Its functions mainly include setting camera parameters, displaying the images captured by the camera, and displaying pseudo-color images. The images captured by the camera can be saved in image format and video format. The UI interface is written through PyQt to communicate with the camera SDK module and the computer UI end (generate solution)

[0108] The fluorescence image enhancement algorithm implemented by this device is a Python implementation based on the Torch framework, which is a tool focused on GPU-accelerated deep neural network programming. Torch is mainly used to process multidimensional matrix data, that is, tensor operations. In PyTorch, common modules include but are not limited to those listed in Table 2, which provide rich functional support for the construction and training of deep learning models.

[0109] Table 2 Commonly used modules in the pytorch framework

[0110]

[0111] The main idea of ​​Retinex-Net is to train a neural network on the collected low-light data blocks. The entire network is divided into two parts, the Decom-Net network is used to achieve image decomposition, and the Enhance-Net network is used to achieve lighting adjustment. In the process of training Decom-Net, there is no GT. When learning the network, it is only necessary to meet the following key constraints, including the consistent reflectivity shared by paired low / normal light images and the consistency of lighting. Enhance-Net is used to achieve lighting enhancement. RetinexNet is an end-to-end low-light enhancement network. A large number of experiments have shown that this method can not only obtain satisfactory low-light enhancement effects, but also can well represent image decomposition. The entire image enhancement process is divided into three steps, including image decomposition, brightness adjustment, and image reconstruction. In the image decomposition stage, the Decom-Net subnetwork decomposes the input image into two parts: a reflectance image and an illumination image. The weights are shared between the two parts, and both input images are decomposed into two parts. In the illumination adjustment stage, an Enhance-Net network based on a codec architecture is used to perform illumination enhancement on the illumination image. This subnetwork contains some residual connections to solve the gradient vanishing problem and performs denoising operations on the reflectance image. In the image reconstruction stage, the illumination image after illumination enhancement and the reflectance image after denoising are recombined to form the final enhanced image.

[0112] In order to reduce the complexity of the model while maintaining good image processing quality, this embodiment proposes a Retinex-Net network structure design with residual connections. Specifically, residual connections are added between the convolutional layers in Decom-Net, so that the network can more easily learn the details and structural information of the image. And residual connections are added between the encoder and decoder of Enhance-Net to improve the learning ability of the network. The specific implementation principle is as follows Figure 3 shown.

[0113] Algorithm processing module design:

[0114] Input the fluorescence image I collected in a low-light environment:

[0115] The input fluorescence image I is decomposed into a reflectance image and an illumination image through the Decom-Net sub-network:

[0116]

[0117] in, and are the estimated reflectance image and illumination image, respectively;

[0118] Shared weights: The network weights are shared between the reflectance image and the illumination image to maintain consistency;

[0119] The brightness of the illuminated image is adjusted and enhanced through the Enhance-Net sub-network based on the codec architecture:

[0120]

[0121] Residual connection: Residual connection is introduced into the network to solve the gradient disappearance problem and improve the efficiency and stability of network training. The residual connection formula is as follows:

[0122]

[0123] Denoising: Denoising is performed on the reflectivity image to eliminate noise in the image and improve image clarity and detail retention:

[0124]

[0125] Image reconstruction: The brightness-adjusted illumination image and the denoised reflectance image are recombined to generate the final enhanced fluorescence image:

[0126]

[0127] Combination strategy: Through the weighted combination strategy, the features of the illumination image and the reflectance image are fused to achieve global brightness adjustment and detail enhancement.

[0128] Before enhancing the collected fluorescence images based on the improved Retinex-Net image enhancement processing algorithm with residual connection, it also includes: constructing a wound bacterial fluorescence image sample data set, and dividing the sample data set into a training set and a test set; using the training set and the test set to train and test a plurality of pre-selected image enhancement processing algorithm models, and using a plurality of evaluation indicators to evaluate the trained plurality of image enhancement processing algorithm models, and obtaining the best image enhancement processing algorithm model according to the evaluation results. The specific contents are as follows:

[0129] 1. Preparation of wound bacterial fluorescence image data

[0130] In order to simulate the wound infection area, a biological tissue phantom with a similar wound size was prepared. Because wound medicine evaluates the bacterial concentration: when the bacterial concentration reaches 106 cfu / ml, the concentration at this time will affect the development of the wound to a worse direction such as ulceration, so this device is equipped with a 106 cfu / ml bacterial solution and collects and processes fluorescent images to observe the imaging effect. The detailed steps are described as follows:

[0131] 1) Add 1 ml of bacterial suspension with a concentration of 109 cfu / ml to 9 ml of saline solution in a 10 ml glass beaker and dilute to make a bacterial solution with a concentration of 106 cfu / ml.

[0132] 2) Place the glass beaker in a vortex oscillator and oscillate for 2 minutes to ensure that the bacteria are evenly distributed and that sufficient sampling is achieved each time.

[0133] 3) Keep the solution at 10 ml during stirring.

[0134] 4) Use a medical plastic pipette to draw 1 ml each time and drop it onto the muscle tissue phantom. Turn on the excitation light source and adjust the angle and distance to illuminate the contaminated area and observe the fluorescence phenomenon.

[0135] 5) Open the host computer software, configure the image acquisition module, and perform a blank experiment to eliminate the interference of fluorescence imaging of the phantom itself.

[0136] 6) Collect fluorescence images of the muscle phantom under different bacterial concentrations and construct a sample data set to provide source data for the subsequent deep learning image enhancement algorithm.

[0137] 2. Comparison of image enhancement results between Retinex algorithm and deep learning algorithm

[0138] The collected fluorescence images were divided into training sets and test sets. The SSR, MSR, MSRCR algorithms, the CNN-based Retinex algorithm and the improved Retinex-Net algorithm were used to train and test the training sets and test sets respectively, and the fluorescence image enhancement model, images and corresponding quality assessment index results were obtained. The indicators used are as follows:

[0139] 1) Structural similarity index measure (SSIM)

[0140]

[0141] Where x and f represent the image blocks of the source image and the fused image in a sliding window respectively. xf represents the covariance of the source image and the fused image, σ x and σ f Represent the standard deviation of the source image and the fused image, μ x and μ f Represent the mean of the source image and the fused image respectively. C1, C2 and C3 are constants used to prevent division by zero.

[0142]

[0143] The SSIM value range is [0,1], and the larger the value, the more similar the images are. If two images are exactly the same, the SSIM value is 1.

[0144] 2) Peak Signal to Noise Ratio (PSNR)

[0145] PSNR is the most common and widely used objective measurement method for evaluating images. Its unit is dB. The larger the value, the less image distortion. Generally speaking, a PSNR higher than 40dB indicates that the image quality is almost the same as the original. Figure 1 A PSNR between 30-40dB usually indicates that the image quality loss is within an acceptable range. A PSNR below 20dB indicates that the image is severely distorted.

[0146] Given a grayscale image I and a noise image K of size m×n, the mean square error (MSE) formula is as follows:

[0147]

[0148] On this basis, PSNR is defined as:

[0149]

[0150] 3) Information Entropy (EN)

[0151]

[0152] Where L is the gray level, P l It is the normalized histogram of the corresponding gray level in the enhanced image. The higher the EN is, the richer the information contained in the enhanced image is.

[0153] 4) Average gradient (AG)

[0154]

[0155] In the formula, The enhanced image has a higher AG, which means it contains richer gradient information.

[0156] 3. Experimental results

[0157] 1) Muscle site test results, such as Figure 4 Shown

[0158] Table 3 Evaluation of bacterial fluorescence image quality in muscle area

[0159]

[0160]

[0161] 2) Tendon site experimental results, such as Figure 5 Shown

[0162] Table 4 Evaluation of bacterial fluorescence image quality in tendon area

[0163]

[0164] From the above implementation, whether it is the muscle tissue part or the tendon tissue part, the SSIM and PSNR values ​​of the SSR and MSR algorithms are far below the range of general quality evaluation, and the effect is very poor; from the perspective of EN and AG, the MSRCR algorithm is more prominent. This algorithm has a strong color restoration, which causes the image to be overexposed and unbalanced; while the Retinex-Net algorithm and the residual connection Retinex-Net algorithm proposed in the present invention are similar in value and close to the original image, with a value of nearly 1. Although the Retinex-Net algorithm reaches 0.66 and 0.86 in value, it is necessary to observe as much as possible. From the perspective of fluorescence, the improved Retinex-Net algorithm of the present invention has better visual effects and clearer contours; the PSNR value reflects the size of the image quality. Generally, images with a PSNR value greater than 30dB have better denoising effects. In this regard, the MSR and MSRCR algorithms are slightly lower than the evaluation criteria, while the two algorithms based on CNN are 47.61 and 50.97dB in the muscle part and 37.07 and 45.82dB in the tendon part. Both values ​​are significantly higher than the evaluation criteria. Although the PSNR value may vary due to human eye observation, from the perspective of the algorithm of the present invention, both the value and the human eye observation are in line with expectations.

[0165] Corresponding to the above disclosed wound surface bacterial autofluorescence detection system, the embodiment of the present invention also discloses a wound surface bacterial autofluorescence detection method, such as Figure 6 As shown, it specifically includes:

[0166] S100, using a laser light source of a specific wavelength to illuminate the wound surface to stimulate the autofluorescence of bacteria on the wound surface;

[0167] S200, collect autofluorescence images of bacteria on the wound surface;

[0168] S300, enhancing the collected fluorescence image by using an improved Retinex-Net image enhancement processing algorithm based on residual connections, so as to perform bacteria detection based on the obtained fluorescence enhanced image.

[0169] Furthermore, collecting the autofluorescence image of the bacteria on the wound surface specifically includes:

[0170] During image acquisition, a dual-band bandpass filter was used to filter out light other than bacterial autofluorescence.

[0171] Furthermore, the collected fluorescence image is enhanced by using an improved Retinex-Net image enhancement processing algorithm based on residual connections, specifically including:

[0172] Image enhancement processing is performed based on a pre-trained convolutional neural network, wherein the convolutional neural network includes a Decom-Net subnetwork and an Enhance-Net subnetwork. The image enhancement processing steps include:

[0173] The input fluorescence image I is decomposed into a reflectance image and an illumination image through the Decom-Net sub-network:

[0174]

[0175] in, and are the estimated reflectance image and illumination image, respectively;

[0176] Shared weights: The network weights are shared between the reflectance image and the illumination image to maintain consistency;

[0177] The brightness of the illuminated image is adjusted and enhanced through the Enhance-Net sub-network based on the codec architecture:

[0178]

[0179] Residual connection: Residual connection is introduced into the network to solve the gradient disappearance problem and improve the efficiency and stability of network training. The residual connection formula is as follows:

[0180]

[0181] Denoising: Denoising is performed on the reflectivity image to eliminate noise in the image and improve image clarity and detail retention:

[0182]

[0183] Image reconstruction: The brightness-adjusted illumination image and the denoised reflectance image are recombined to generate the final enhanced fluorescence image:

[0184]

[0185] Combination strategy: Through the weighted combination strategy, the features of the illumination image and the reflectance image are fused to achieve global brightness adjustment and detail enhancement.

[0186] Furthermore, before enhancing the collected fluorescence image based on the improved Retinex-Net image enhancement processing algorithm with residual connection, the method further includes:

[0187] Constructing a wound surface bacterial fluorescence image sample dataset, and dividing the sample dataset into a training set and a test set;

[0188] The training set and the test set are used to train and test a plurality of pre-selected image enhancement processing algorithm models, and a plurality of evaluation indicators are used to evaluate the trained plurality of image enhancement processing algorithm models, and the best image enhancement processing algorithm model is obtained according to the evaluation results.

[0189] Furthermore, the pre-selected multiple image enhancement processing algorithm models include SSR, MSR, MSRCR algorithms, Retinex-Net algorithm and improved Retinex-Net algorithm with residual connection;

[0190] The evaluation indicators include structural similarity measurement SSIM, peak signal-to-noise ratio PSNR, information entropy EN, and average gradient AG.

[0191] Furthermore, a wound bacterial fluorescence image sample dataset is constructed, specifically including:

[0192] 1 ml of bacterial suspension with a concentration of 109 cfu / ml was mixed with 9 ml of saline solution in a 10 ml glass beaker to dilute and make a bacterial solution with a concentration of 106 cfu / ml.

[0193] Place the glass beaker in a vortex oscillator and oscillate for 2 minutes to evenly distribute the bacteria and ensure that sufficient sampling is done each time. During the stirring process, keep the solution at 10 ml.

[0194] Use a medical plastic pipette to draw 1 ml each time and drop it onto the biological tissue phantom, turn on the excitation light source and adjust the angle and distance to illuminate the contaminated area to observe the fluorescence phenomenon;

[0195] Open the host computer software, configure the image acquisition module, and conduct a blank experiment to eliminate the interference of fluorescence imaging of the biological tissue phantom itself;

[0196] Fluorescence images of biological tissue phantoms under different bacterial concentrations were collected to construct a sample data set.

[0197] It should be noted that for a detailed description of a wound bacterial autofluorescence detection method provided in an embodiment of the present invention, reference can be made to the relevant description of a wound bacterial autofluorescence detection system provided in an embodiment of the present application, which will not be repeated here.

[0198] The above specific examples are used to illustrate the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the art, according to the concept of the present invention, some simple deductions, modifications or substitutions can be made.

Claims

1. A wound bacterial autofluorescence detection system, characterized in that: The system includes an excitation light source module, an image acquisition module and a host computer; The excitation light source module is used to use a laser light source of a specific wavelength to irradiate the wound surface to stimulate the autofluorescence of bacteria on the wound surface; The image acquisition module is used to acquire the autofluorescence image of the bacteria on the wound surface and send the acquired fluorescence image to the host computer; The host computer is used to enhance the collected fluorescence image based on the improved Retinex-Net image enhancement processing algorithm with residual connection, so as to perform bacteria detection based on the obtained fluorescence enhanced image; The collected fluorescence images are enhanced based on the improved Retinex-Net image enhancement processing algorithm with residual connection, including: Image enhancement processing is performed based on a pre-trained convolutional neural network, wherein the convolutional neural network includes a Decom-Net subnetwork and an Enhance-Net subnetwork. The image enhancement processing steps include: The input fluorescence image I is decomposed into a reflectance image and an illumination image through the Decom-Net sub-network: in, and are the estimated reflectance image and illumination image, respectively; Shared weights: The network weights are shared between the reflectance image and the illumination image to maintain consistency; The brightness of the illuminated image is adjusted and enhanced through the Enhance-Net sub-network based on the codec architecture: Residual connection: Residual connection is introduced into the network to solve the gradient disappearance problem and improve the efficiency and stability of network training. The residual connection formula is as follows: Denoising: Denoising is performed on the reflectivity image to eliminate noise in the image and improve image clarity and detail retention: Image reconstruction: The brightness-adjusted illumination image and the denoised reflectance image are recombined to generate the final enhanced fluorescence image: Combination strategy: Through the weighted combination strategy, the features of the illumination image and the reflectance image are fused to achieve global brightness adjustment and detail enhancement.

2. A wound surface bacterial autofluorescence detection system as claimed in claim 1, characterized in that: The excitation light source module includes an excitation light source and a light source driving module connected to the excitation light source; The excitation light source uses LED lamp beads with a peak wavelength of a preset value; The irradiation radius of a single lamp bead is: r = tanθ × d Where θ is the half angle of the light source divergence angle, d is the distance from the illuminated surface to the light source, and the illuminated area of ​​the illuminated surface is expressed as: S = π × (tanθ × d)2 Where S is the area of ​​the surface irradiated by the light source. When the radiation flux of the light source is known, The radiation flux or irradiance of the incident surface is expressed as: The light source driving module includes a power chip and a constant current driver, and changes the brightness of the light source by controlling the output power.

3. A wound surface bacterial autofluorescence detection system as claimed in claim 1, characterized in that: The image acquisition module includes a camera and a dual-band bandpass filter; The camera is used to collect autofluorescence images of bacteria on the wound surface. When shooting, the camera is 10-30 cm away from the wound surface, and the lens is a zoom lens with a focal length of 8-50 mm; The dual-band bandpass filter is used to filter out light in bands other than bacterial autofluorescence during image acquisition.

4. A method for detecting wound bacterial autofluorescence, characterized in that: The method comprises: A laser light source with a specific wavelength is used to illuminate the wound surface to stimulate the autofluorescence of bacteria on the wound surface; Collect autofluorescence images of bacteria on the wound surface; The collected fluorescence images are enhanced by using an improved Retinex-Net image enhancement processing algorithm based on residual connections, so as to detect bacteria based on the obtained fluorescence enhanced images; The acquired fluorescence images are enhanced by using an improved Retinex-Net image enhancement processing algorithm based on residual connections, specifically including: Image enhancement processing is performed based on a pre-trained convolutional neural network, wherein the convolutional neural network includes a Decom-Net subnetwork and an Enhance-Net subnetwork. The image enhancement processing steps include: The input fluorescence image I is decomposed into a reflectance image and an illumination image through the Decom-Net sub-network: in, and are the estimated reflectance image and illumination image, respectively; Shared weights: The network weights are shared between the reflectance image and the illumination image to maintain consistency; The brightness of the illuminated image is adjusted and enhanced through the Enhance-Net sub-network based on the codec architecture: Residual connection: Residual connection is introduced into the network to solve the gradient disappearance problem and improve the efficiency and stability of network training. The residual connection formula is as follows: Denoising: Denoising is performed on the reflectivity image to eliminate noise in the image and improve image clarity and detail retention: Image reconstruction: The brightness-adjusted illumination image and the denoised reflectance image are recombined to generate the final enhanced fluorescence image: Combination strategy: Through the weighted combination strategy, the features of the illumination image and the reflectance image are fused to achieve global brightness adjustment and detail enhancement.

5. A wound surface bacterial autofluorescence detection method according to claim 4, characterized in that: Collect autofluorescence images of bacteria on the wound surface, including: During image acquisition, a dual-band bandpass filter was used to filter out light other than bacterial autofluorescence.

6. A wound surface bacterial autofluorescence detection method as claimed in claim 4, characterized in that: Before enhancing the acquired fluorescence image based on the improved Retinex-Net image enhancement processing algorithm with residual connection, the following is also included: Constructing a wound surface bacterial fluorescence image sample dataset, and dividing the sample dataset into a training set and a test set; The training set and the test set are used to train and test a plurality of pre-selected image enhancement processing algorithm models, and a plurality of evaluation indicators are used to evaluate the trained plurality of image enhancement processing algorithm models, and the best image enhancement processing algorithm model is obtained according to the evaluation results.

7. A wound bacterial autofluorescence detection method as claimed in claim 6, characterized in that: The pre-selected image enhancement processing algorithm models include SSR, MSR, MSRCR algorithm, Retinex-Net algorithm and improved Retinex-Net algorithm with residual connection; The evaluation indicators include structural similarity measurement SSIM, peak signal-to-noise ratio PSNR, information entropy EN, and average gradient AG.

8. A wound surface bacterial autofluorescence detection method as claimed in claim 6, characterized in that: Construct a sample dataset of wound bacterial fluorescence images, including: 1 ml of bacterial suspension with a concentration of 109 cfu / ml was mixed with 9 ml of saline solution in a 10 ml glass beaker to dilute and make a bacterial solution with a concentration of 106 cfu / ml. Place the glass beaker in a vortex oscillator and oscillate for 2 minutes to evenly distribute the bacteria and ensure that sufficient sampling is done each time. During the stirring process, keep the solution at 10 ml. Use a medical plastic pipette to draw 1 ml each time and drop it onto the biological tissue phantom, turn on the excitation light source and adjust the angle and distance to illuminate the contaminated area to observe the fluorescence phenomenon; Open the host computer software, configure the image acquisition module, and conduct a blank experiment to eliminate the interference of fluorescence imaging of the biological tissue phantom itself; Fluorescence images of biological tissue phantoms under different bacterial concentrations were collected to construct a sample data set.

Images