Pathological tissue area enhanced identification method based on skin disease dynamic epidermal response
By constructing cell proliferation weights and adaptive cropping coefficient adjustments in psoriasis pathological images, combined with PWC-Net and CLAHE algorithms, the problem of low accuracy in psoriasis pathological tissue recognition in the prior art is solved, and efficient enhancement and accurate recognition of pathological images are achieved.
Patent Information
- Application Number
- CN202510983972.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-17
- Publication Date
- 2025-08-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art is difficult to effectively capture the timing information of basal cell proliferation and vertical differentiation and migration changes in epidermal dynamic reactions during the identification of psoriasis pathological tissues, resulting in blurred lesion boundaries in the pathological image and poor enhancement effect of multiple types of cell mixed regions, and thus low recognition accuracy.
By obtaining pathological images of different time nodes, combining the local region timing change characteristics of the epidermal dynamic reaction process, cell proliferation weights are constructed, and the pathological images are enhanced by adaptive cropping coefficient adjustment method. The optical flow vector is extracted using the PWC-Net network, combined with the CLAHE algorithm for image enhancement, and the U-Net++ model is used to segment and identify pathological tissue regions.
It significantly improves the enhancement effect of pathological images, highlights the characteristics of pathological tissue areas, and improves the identification accuracy of pathological tissue areas. Especially in the identification of psoriasis pathological tissue areas, the contrast of the lesion area is enhanced and the recognition accuracy is improved.
Smart Images
Figure CN120495296A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of image processing technology, and in particular to a method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases. Background Art
[0002] The dynamic response of the epidermis of skin diseases (including dynamic changes such as cell proliferation, inflammatory infiltration, and abnormal keratinization) is an intuitive physiological mapping of the pathological tissue area. Its temporal and spatial characteristics (such as the evolution process of cell morphology, dynamic characteristics of blood vessel distribution, and degree of tissue structural disorder) are closely related to the boundary definition, type classification, and severity of pathological tissue.
[0003] With the development of artificial intelligence, machine learning methods are currently used to identify pathological areas and categories in pathological images in research that combines pathological tissue area recognition with epidermal dynamic responses. Pathological images usually need to be enhanced before recognition. The most significant problem at present is that in the process of identifying pathological tissue areas for psoriasis, traditional methods have difficulty effectively capturing temporal information such as basal cell proliferation and vertical differentiation and migration changes in epidermal dynamic responses when enhancing pathological images. This leads to blurred lesion boundaries in pathological images and poor enhancement effects in areas where multiple cell types are mixed, resulting in low accuracy in identifying pathological tissue areas in images. Summary of the Invention
[0004] In order to solve the above technical problems, the present application provides a pathological tissue area enhanced identification method based on the dynamic epidermal response of skin diseases to solve the existing problems.
[0005] The present invention discloses a method for enhanced identification of pathological tissue regions based on dynamic epidermal reactions of skin diseases using the following technical solutions: One embodiment of the present application provides a method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases, the method comprising the following steps: Acquire all pathological images in each acquisition cycle, and the fluorescently marked area in each pathological image; The last frame of the pathological image in each acquisition cycle is recorded as the image to be enhanced in each acquisition cycle; based on the optical flow vectors of each pixel point between each adjacent frame image in the acquisition cycle of each image to be enhanced, the optical flow vector set of each pixel point in each image to be enhanced is obtained; based on the modulus value of each vector in the optical flow vector set of each pixel point, the displacement cumulative characteristic value of each pixel point is obtained; all neighboring points of each pixel point in each image to be enhanced are obtained, and based on the similarity between the optical flow vector set of each pixel point in each image to be enhanced and its each neighboring point, and the difference between the displacement cumulative characteristic values between each pixel point and its each neighboring point, the dynamic saliency value of each pixel point in each image to be enhanced is obtained; A sliding window is set in each image to be enhanced; the cell proliferation weight of each sliding window in each image to be enhanced is obtained based on the proportion of pixels belonging to the fluorescent labeled area in each sliding window in each image to be enhanced, the displacement cumulative characteristic value of each pixel in each sliding window, and the dynamic significance value; the adjustment weight of each pixel is obtained based on the average level of the cell proliferation weights of all sliding windows corresponding to each pixel in the image to be enhanced; each image to be enhanced is evenly divided into multiple enhancement regions, and the cropping coefficient of each enhancement region in each image to be enhanced is obtained based on the average level of the adjustment weights of all pixels in each enhancement region, and then each image to be enhanced is enhanced.
[0006] Preferably, the process of acquiring the set of optical flow vectors of each pixel point in each image to be enhanced is as follows: starting from the second frame of pathological image in each acquisition cycle, each pathological image and its previous frame image are respectively used as the input of the PWC-Net network, and the optical flow vector of each pixel point in each pathological image is output; the optical flow vectors of each pixel point in each image to be enhanced and the optical flow vectors of the corresponding pixels at the same position in all other pathological images in the acquisition cycle are respectively set in time sequence, and recorded as the set of optical flow vectors of each pixel point in each image to be enhanced.
[0007] Preferably, the cumulative displacement characteristic value of each pixel point refers to the sum of the module values of all optical flow vectors in the optical flow vector set of each pixel point.
[0008] Preferably, the process of acquiring the dynamic saliency value of each pixel in each image to be enhanced is: Obtaining a first eigenvalue between each pixel point in each to-be-enhanced image and each of its neighboring points according to a similarity between a set of optical flow vectors of each pixel point in each to-be-enhanced image and each of its neighboring points; Calculate the dynamic saliency value of each pixel in each image to be enhanced: Where, Indicates the first The dynamic saliency value of each pixel, Indicates the first The cumulative characteristic value of the displacement of the pixel point, Indicates the The first pixel The cumulative displacement eigenvalue of the neighboring points is Indicates the first Pixels and their The first eigenvalue between the nearest neighbor points, Indicates a preset constant, Indicates the first The total number of neighboring points of a pixel.
[0009] Preferably, the first eigenvalue between each pixel point in each image to be enhanced and its neighboring points refers to the mean of the cosine similarities between the set of optical flow vectors of each pixel point in each image to be enhanced and all optical flow vectors of the same order in the set of optical flow vectors of its neighboring points.
[0010] Preferably, the specific process of setting a sliding window in each image to be enhanced is: setting a sliding window of size w×w in each image to be enhanced, sliding with a preset sliding step length e, wherein w is the preset side length of the sliding window, and e <w。
[0011] Preferably, the calculation formula for the cell proliferation weight of each sliding window in each image to be enhanced is: Where, represents the cell value-added weight of the j-th sliding window in the image to be enhanced, represents the density of Ki-67 positive cells in the jth sliding window in the image to be enhanced, Indicates the jth sliding window in the image to be enhanced The cumulative characteristic value of the displacement of the pixel point, Indicates the jth sliding window in the image to be enhanced The dynamic saliency value of each pixel, represents the sum of the dynamic saliency values of all pixels in the j-th sliding window in the image to be enhanced, represents the total number of pixels in the j-th sliding window in the image to be enhanced; wherein, the density of Ki-67-positive cells in the j-th sliding window refers to the proportion of the total number of pixels in the fluorescent-labeled area among all pixels in the j-th sliding window.
[0012] Preferably, the adjustment weight of each pixel point refers to the average of the normalized cell proliferation weights of all sliding windows corresponding to each pixel point.
[0013] Preferably, the calculation formula for the cropping coefficient of each enhanced area in each image to be enhanced is: Where, represents the cropping coefficient of the u-th enhanced region in the image to be enhanced; Indicates the preset adjustment factor, whose value range is [0.3,0.5]; It represents the mean of the adjustment weights of all pixels in the u-th enhancement area in the image to be enhanced.
[0014] Preferably, the specific process of enhancing each image to be enhanced is: taking all enhanced areas in each image to be enhanced and their corresponding cropping coefficients as inputs of the CLAHE algorithm, enhancing all enhanced areas in each image to be enhanced, and obtaining enhanced images to be enhanced.
[0015] This application has at least the following beneficial effects: Because psoriasis shows significant differences in tissue regional characteristics at different stages of the disease course, it is difficult to effectively capture temporal information such as basal cell proliferation and vertical differentiation and migration changes in the dynamic response of the epidermis, resulting in low recognition accuracy of traditional pathological tissue region recognition methods in the process of psoriasis pathological tissue region recognition; this application collects pathological images of lesion areas from different patients, and for pathological images obtained at different time nodes, combines the temporal change characteristics of local areas in the epidermal dynamic response process, deeply analyzes the dynamic epidermal response characteristics, constructs cell proliferation weights, and then assigns different cropping coefficients to images in different areas, thereby improving the enhancement effect of pathological images and effectively highlighting the pathological tissue region characteristics in the image; its core advantage lies in closely combining the local temporal change characteristics of basal cell proliferation and vertical differentiation and migration in different stages of psoriasis, accurately analyzing the dynamic epidermal response of the lesion area, and significantly improving the contrast of the pathological tissue area in the image through regional optimization and adjustment of the enhancement strategy, thereby improving the image enhancement effect of the pathological image, which is conducive to improving the subsequent accuracy of pathological tissue region recognition. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0017] Figure 1 A flowchart of the steps of the method for enhanced identification of pathological tissue areas based on dynamic epidermal reactions of skin diseases provided in this application; Figure 2 This is a flowchart for obtaining the cropping coefficients of each enhanced area in the image to be enhanced provided by this application. DETAILED DESCRIPTION
[0018] In order to further illustrate the technical means and effects adopted by this application to achieve the predetermined invention objectives, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features and effects of the pathological tissue area enhanced identification method based on the dynamic epidermal reaction of skin diseases proposed in this application. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics of one or more embodiments may be combined in any suitable form.
[0019] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
[0020] The specific scheme of the pathological tissue region enhanced identification method based on the dynamic epidermal reaction of skin diseases provided by the present application is described in detail below with reference to the accompanying drawings.
[0021] An embodiment of the present application provides a method for enhancing the recognition of pathological tissue regions based on the dynamic epidermal reaction of skin diseases. Specifically, the following method for enhancing the recognition of pathological tissue regions based on the dynamic epidermal reaction of skin diseases is provided. Figure 1 , the method comprises the following steps: Step 1: Acquire all pathological images in each acquisition cycle and the fluorescent marked areas in each pathological image.
[0022] At key time points, such as before, during, and after treatment, Ki-67-positive cells in the lesion area were labeled using fluorescence signals. Multiple scans of the same lesion were performed using in vivo confocal laser scanning microscopy (IVCM). An acquisition cycle of N minutes was defined, with N ranging from 10 to 20, and in this example, N was 15. The acquisition frequency was 1 Hz. Pathological images capturing the dynamic processes of cell proliferation and migration were acquired at a depth of 200 μm below the epidermis, enabling clear observation of the distribution and changes of Ki-67-positive cells (a cell proliferation marker). Because fluorescence signals are susceptible to background noise, a median filter algorithm was used in this example to eliminate isolated noise points in the pathological images, highlighting the fluorescence signals of Ki-67-positive cells. The fluorescence signals in the pathological images were then used to count the fluorescently labeled areas.
[0023] Step 2: The last frame of pathological image in each acquisition cycle is recorded as the image to be enhanced in each acquisition cycle; based on the optical flow vector of each pixel point between each adjacent frame image in the acquisition cycle of each image to be enhanced, the optical flow vector set of each pixel point in each image to be enhanced is obtained; based on the modulus value of each vector in the optical flow vector set of each pixel point, the displacement cumulative eigenvalue of each pixel point is obtained; all neighboring points of each pixel point in each image to be enhanced are obtained, and based on the similarity between the optical flow vector set of each pixel point in each image to be enhanced and its neighboring points, and the difference in the displacement cumulative eigenvalue between each pixel point and its neighboring points, the dynamic saliency value of each pixel point in each image to be enhanced is obtained.
[0024] Psoriasis exhibits significant dynamic epidermal reactions at different stages of the disease course: The progressive stage is characterized by rapid thickening of the epidermal spinous layer (approximately 5-10 μm per day), increased downward extension and branching of epidermal processes, and a sharp increase in the number of Munro microabscesses (up to 10-20 per square millimeter). During the stable stage, epidermal thickening slows and microabscesses gradually decrease. During the resolving stage, epidermal thickness returns to normal, and inflammatory cell infiltration significantly decreases. Based on these dynamic characteristics, pathological image enhancement can be optimized by calculating cell proliferation weights to highlight the characteristics of pathological tissue regions in the image.
[0025] Specifically, based on all denoised pathology images within a single acquisition cycle, starting with the second frame, each pathology image and its previous frame are used as input to the PWC-Net network. This network extracts features at multiple scales by constructing an image pyramid and uses deformable convolution to capture the complex trajectories of cell movement. It outputs an optical flow vector (including displacement magnitude and direction) for each pixel in each pathology image, reflecting the temporal movement of cells in the image. It should be noted that this process calculates the optical flow vector for each pixel in each frame using the previous frame as a reference, thereby obtaining the optical flow vector for each pixel in each frame except the first frame within a single acquisition cycle.
[0026] Given that the dynamic response characteristics of the epidermis vary across different stages of psoriasis, it is necessary to analyze the dynamic epidermal response characteristics in combination with the temporal motion changes of pixels within a local area to enhance the pathological tissue characteristics of different stages and regions. The last pathological image frame within each acquisition cycle is recorded as the image to be enhanced for each acquisition cycle. The following analysis uses the image to be enhanced within a single acquisition cycle as an example. The optical flow vectors of each pixel in the image to be enhanced within a single acquisition cycle and the optical flow vectors of the corresponding pixels at the same position in all previous pathological images are collected in temporal order and recorded as the optical flow vector set of each pixel in the image to be enhanced within a single acquisition cycle.
[0027] To address the cellular movement characteristics of epidermal branching during the advanced phase (e.g., abnormal proliferation, polarity disturbance, and collective migration of basal keratinocytes, distinct from the unidirectional vertical differentiation and migration of normal epidermis), the local dynamic variation of each pixel in the enhanced image is calculated based on the optical flow vector set. First, the cosine similarity between the optical flow vector set of each pixel in the enhanced image and all optical flow vectors of the same position in the optical flow vector set of its i-th neighbor is calculated, using the eight neighboring pixels of each pixel as its nearest neighbors. This mean is recorded as the first eigenvalue between each pixel and its i-th neighbor. A smaller first eigenvalue indicates a greater difference in the dynamic variation between the corresponding pixel and its neighbors during the acquisition period. This indicates a greater likelihood that the region containing the pixel is experiencing an increase in branching due to basal cell proliferation, and a more significant difference from normal vertical differentiation and migration.
[0028] Furthermore, in order to accurately highlight the changing characteristics of the pathological tissue area due to dynamic epidermal reactions during the pathological tissue area identification process, the sum of the moduli of all optical flow vectors in the optical flow vector set of each pixel point in the image to be enhanced is used as the displacement cumulative eigenvalue of each pixel point. The larger the displacement cumulative eigenvalue, the more significant the cell movement change characteristics in the tissue area caused by base cell proliferation.
[0029] As a preferred embodiment, the dynamic saliency value of each pixel point in each image to be enhanced is obtained based on the similarity between the optical flow vector sets of each pixel point and its neighboring points in each image to be enhanced, as well as the difference in the cumulative displacement characteristic values between each pixel point and its neighboring points, to characterize the saliency of the dynamic epidermal reaction characteristics at the location of each pixel point in the image to be enhanced.
[0030] In this embodiment, the first The dynamic saliency value of a pixel is recorded as , its specific expression is: Where, Indicates the first The dynamic saliency value of each pixel, Indicates the first The cumulative characteristic value of the displacement of the pixel point, Indicates the The first pixel The cumulative displacement eigenvalue of the neighboring points is Indicates the first Pixels and their The first eigenvalue between neighboring points; represents a preset constant, which is 0.01 in this embodiment; Indicates the first The total number of neighboring points of each pixel. In this embodiment, n is 8.
[0031] The larger the calculated dynamic significance value is, the more it represents the cumulative displacement characteristics of the dynamic epidermal response of psoriasis at different stages and the differences in vertical differentiation migration caused by the increase in extended branches; the more significant the dynamic epidermal response characteristics at the position of the current pixel are.
[0032] Step 3: Set sliding windows in each image to be enhanced; obtain the cell proliferation weights of each sliding window in each image to be enhanced according to the proportion of pixel points belonging to the fluorescent labeled area in each sliding window, the cumulative displacement characteristic values and dynamic significance values of each pixel point in each sliding window; obtain the adjustment weights of each pixel point according to the average level of the cell proliferation weights of all sliding windows corresponding to each pixel point in the image to be enhanced; evenly divide each image to be enhanced into multiple enhancement regions, and obtain the cropping coefficients of each enhancement region in each image to be enhanced according to the average level of the adjustment weights of all pixel points in each enhancement region, and then enhance each image to be enhanced.
[0033] Through the above steps, a comprehensive analysis is carried out on the significance of the dynamic epidermal response characteristics of the pathological tissue area caused by basal cell proliferation in different regions. Compared with the identification of the psoriasis pathological tissue area, due to the differences in the pathological characteristics presented by psoriasis at different stages, misjudgment may occur during the identification process. Therefore, the purpose of enhancing and adjusting the actually acquired image in combination with the dynamic epidermal characteristics is to highlight the characteristics of the psoriasis pathological tissue area and improve the identification accuracy.
[0034] Specifically, considering that the differences in the dynamic epidermal response characteristics within the pathological tissue area are mainly affected by cell proliferation and the differences in vertical differentiation migration accompanying cell proliferation, this application sets sliding windows for local area analysis in the image to be enhanced. In an implementation process of this application, to avoid a significant increase in local feature extraction errors caused by an overly large sliding window, the size of the sliding window is set to w×w and it slides with a preset sliding step e (in other implementation manners, the window size can be adaptively adjusted according to the actual situation). Here, w is the preset side length of the sliding window, and to fully reflect the local dynamic characteristics of different pixel points, let e < w, that is, there is an overlapping area between the windows before and after sliding. In this embodiment, w is taken as 21 and e is taken as 10. Count the total number of pixel points belonging to the fluorescent labeled area in each sliding window in the image to be enhanced, and take the proportion of the total number of pixel points belonging to the fluorescent labeled area among all pixel points in each sliding window as the density of Ki-67 positive cells in each sliding window.
[0035] As a preferred embodiment, the cell proliferation weight of each sliding window in each image to be enhanced is obtained based on the proportion of pixels belonging to the fluorescent marked area in each sliding window in each image to be enhanced, the cumulative characteristic value of the displacement of each pixel in each sliding window, and the dynamic significance value, which is used to characterize the significance of the pathological tissue characteristics of the area where each sliding window in each image to be enhanced is located.
[0036] In this embodiment, the cell increment weight of the jth sliding window in the image to be enhanced is recorded as , its specific expression is: Where, represents the cell value-added weight of the j-th sliding window in the image to be enhanced, represents the density of Ki-67 positive cells in the jth sliding window in the image to be enhanced, Indicates the jth sliding window in the image to be enhanced The cumulative characteristic value of the displacement of the pixel point, Indicates the jth sliding window in the image to be enhanced The dynamic saliency value of each pixel, represents the sum of the dynamic saliency values of all pixels in the j-th sliding window in the image to be enhanced, Represents the total number of pixels in the j-th sliding window in the image to be enhanced.
[0037] If the Ki-67 positive cells in the area where the sliding window is located in the enhanced image proliferate more, the cells are more active during proliferation, and the dynamic epidermal reaction characteristics of the proliferation process are more significant, then the pathological tissue characteristics in the area where the sliding window is located are more significant, and the calculated cell proliferation weight is greater.
[0038] Furthermore, all pathological images within a single acquisition cycle can record the dynamic changes of the psoriatic epidermis and reflect activities such as cell proliferation and migration. Therefore, the cell proliferation weights of each sliding window in the image to be enhanced contain rich dynamic information features.
[0039] This application uses the CLAHE (Contrast-Limited Adaptive Histogram Equalization) algorithm for image enhancement. When using the CLAHE algorithm for image enhancement, the value of the cropping factor plays a key role in the image enhancement effect. When the cropping factor is set to a small value, the image contrast is slightly improved; when it is set to a large value, the image contrast is greatly improved. Therefore, based on the cell proliferation weight of the image to be enhanced, the cropping factor of different regions in the image to be enhanced can be optimized and adjusted during image enhancement to highlight the pathological features of psoriasis in the image to be enhanced.
[0040] Specifically, the normalized cell proliferation weights of all sliding windows in the image to be enhanced are used as the proliferation change weights for all pixels within the corresponding sliding windows. Because sliding windows overlap, meaning each pixel may fall within multiple sliding windows, a single pixel may correspond to multiple proliferation change weights. To fully reflect the significance of the local dynamic information characteristics of different pixels, the average of all proliferation change weights corresponding to each pixel is used as the adjustment weight for each pixel. Furthermore, based on the adjustment weights of each pixel in each acquired image to be enhanced, the cropping coefficients for different regions in the CLAHE algorithm are adjusted.
[0041] In view of the differences in the dynamic epidermal reaction characteristics at different stages of psoriasis, when using the CLAHE algorithm for image enhancement, the image to be enhanced is divided into The image is divided into multiple enhancement regions according to the size of the image to be enhanced, and then the cropping coefficient of each enhancement region is calculated. In this embodiment, the cropping coefficient of the u-th enhancement region in the image to be enhanced is recorded as , its specific expression is: Where, represents the cropping coefficient of the u-th enhanced region in the image to be enhanced; Represents the preset adjustment factor, whose value range is [0.3, 0.5]. In this embodiment, it is set to 0.4. Its purpose is to prevent excessive enhancement during the adjustment process from causing image distortion or noise amplification; Represents the mean of the adjustment weights of all pixels in the u-th enhanced region in the image to be enhanced. The flow chart for obtaining the cropping coefficients of each enhanced region in the image to be enhanced is as follows: Figure 2 shown.
[0042] The larger the mean of the adjustment weights in the enhanced region is, the more likely it is that the enhanced region contains psoriasis pathological features. Therefore, in order to highlight the psoriasis pathological features in the region, a larger clipping coefficient is set so that the contrast in the enhanced region is greatly improved.
[0043] It should be noted that in the process of identifying pathological tissue areas in psoriasis, it is difficult to effectively capture temporal information such as basal cell proliferation and vertical differentiation and migration changes in the dynamic response of the epidermis, resulting in poor enhancement processing effects in areas with blurred lesion boundaries and mixed cell types. Compared with the traditional method of determining and adjusting the cropping coefficient, the above-mentioned determination method fully considers the basal cell proliferation and vertical differentiation and migration characteristics in different stages of psoriasis, analyzes the dynamic epidermal reaction characteristics, and then enhances and optimizes the collected pathological images based on the analysis results, which can significantly deepen the contrast of the active proliferation area and form a stronger contrast with other areas, thereby highlighting the regional characteristics of the pathological tissue.
[0044] According to the above-mentioned cropping coefficient calculation method, the cropping coefficients of all enhanced areas in the image to be enhanced are calculated, and each enhanced area and its corresponding cropping coefficient are used as the input of the CLAHE algorithm. Each enhanced area in the image to be enhanced is enhanced to obtain an enhanced image to be enhanced, and the enhanced image to be enhanced is recorded as the enhanced image.
[0045] After the above adjustments, the contrast of cell proliferation and active movement areas in pathological images is enhanced.
[0046] Furthermore, a deep learning model is used to segment and identify the enhanced pathological images. Considering that the U-Net++ model enhances the ability to integrate features at different levels by introducing nested skip connections, it is more suitable for the segmentation of complex pathological structures. Therefore, this application adopts the U-Net++ model for the segmentation and identification of pathological tissue areas.
[0047] During the model training process, this embodiment obtains pathological images of 500 acquisition cycles in total, and obtains 500 enhanced images after the image to be enhanced in each acquisition cycle is subjected to the above-mentioned enhancement process. All enhanced images are divided into a training set and a test set in a ratio of 7:3 to obtain a pathological image dataset; the pathological image dataset is used as the input of the U-Net++ model, and the model parameters are optimized by combining the Dice loss function with the cross entropy loss function. The Adam optimizer is used, the initial learning rate is set to 0.001, and it decays by 10% every 10 epochs to obtain a pathological tissue recognition model. The obtained pathological tissue recognition model can be used to identify and classify the pathological tissue areas in the pathological images to be identified, where the classification results include background, hyperkeratosis layer, parakeratosis layer, spinous layer / granular layer and epidermal protrusions / dermal papillae. The specific model training process is well known to those skilled in the art and will not be described in detail here.
[0048] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0049] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.
[0050] The above-described embodiments are only used to illustrate the technical solutions of the present application, and not to limit them. Modifications to the technical solutions described in the aforementioned embodiments, or equivalent replacements of some of the technical features therein, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application, and should all be included in the scope of protection of the present application.
Claims
1. A pathological tissue region enhancement recognition method based on dynamic epidermal response of skin diseases, characterized in that: The method comprises the following steps: Acquire all pathological images in each acquisition cycle, and the fluorescently marked area in each pathological image; The last frame of the pathological image in each acquisition cycle is recorded as the image to be enhanced in each acquisition cycle; based on the optical flow vectors of each pixel point between each adjacent frame image in the acquisition cycle of each image to be enhanced, the optical flow vector set of each pixel point in each image to be enhanced is obtained; based on the modulus value of each vector in the optical flow vector set of each pixel point, the displacement cumulative characteristic value of each pixel point is obtained; all neighboring points of each pixel point in each image to be enhanced are obtained, and based on the similarity between the optical flow vector set of each pixel point in each image to be enhanced and its each neighboring point, and the difference between the displacement cumulative characteristic values between each pixel point and its each neighboring point, the dynamic saliency value of each pixel point in each image to be enhanced is obtained; A sliding window is set in each image to be enhanced; the cell proliferation weight of each sliding window in each image to be enhanced is obtained based on the proportion of pixels belonging to the fluorescent labeled area in each sliding window in each image to be enhanced, the displacement cumulative characteristic value of each pixel in each sliding window, and the dynamic significance value; the adjustment weight of each pixel is obtained based on the average level of the cell proliferation weights of all sliding windows corresponding to each pixel in the image to be enhanced; each image to be enhanced is evenly divided into multiple enhancement regions, and the cropping coefficient of each enhancement region in each image to be enhanced is obtained based on the average level of the adjustment weights of all pixels in each enhancement region, and then each image to be enhanced is enhanced.
2. The method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases according to claim 1, characterized in that: The process of acquiring the optical flow vector set of each pixel point in each image to be enhanced is as follows: starting from the second frame of pathological image in each acquisition cycle, each pathological image and its previous frame image are respectively used as the input of the PWC-Net network, and the optical flow vector of each pixel point in each pathological image is output; the optical flow vector of each pixel point in each image to be enhanced and the optical flow vectors of the corresponding pixels at the same position in all other pathological images in the acquisition cycle are respectively set in time sequence, and recorded as the optical flow vector set of each pixel point in each image to be enhanced.
3. The method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases according to claim 1, characterized in that: The cumulative displacement characteristic value of each pixel point refers to the sum of the module values of all optical flow vectors in the optical flow vector set of each pixel point.
4. The method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases according to claim 1, characterized in that: The process of obtaining the dynamic saliency value of each pixel in each image to be enhanced is as follows: Obtaining a first eigenvalue between each pixel point in each to-be-enhanced image and each of its neighboring points according to a similarity between a set of optical flow vectors of each pixel point in each to-be-enhanced image and each of its neighboring points; Calculate the dynamic saliency value of each pixel in each image to be enhanced: Where, Indicates the first The dynamic saliency value of each pixel, Indicates the first The cumulative characteristic value of the displacement of the pixel point, Indicates the The first pixel The cumulative displacement eigenvalue of the neighboring points is Indicates the first Pixels and their The first eigenvalue between the nearest neighbor points, Indicates a preset constant, Indicates the first The total number of neighboring points of a pixel.
5. The method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases according to claim 4, characterized in that: The first eigenvalue between each pixel point in each image to be enhanced and its neighboring points refers to the average of the cosine similarities between the set of optical flow vectors of each pixel point in each image to be enhanced and all optical flow vectors of the same order in the set of optical flow vectors of its neighboring points.
6. The method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases according to claim 1, characterized in that: The specific process of setting the sliding window in each image to be enhanced is as follows: setting a sliding window of size w×w in each image to be enhanced, and sliding with a preset sliding step length e, wherein w is the preset side length of the sliding window, and e <w。 7. The method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases according to claim 1, characterized in that: The calculation formula of the cell proliferation weight of each sliding window in each image to be enhanced is: Where, represents the cell value-added weight of the j-th sliding window in the image to be enhanced, represents the density of Ki-67 positive cells in the jth sliding window in the image to be enhanced, Indicates the jth sliding window in the image to be enhanced The cumulative characteristic value of the displacement of the pixel point, Indicates the jth sliding window in the image to be enhanced The dynamic saliency value of each pixel, represents the sum of the dynamic saliency values of all pixels in the j-th sliding window in the image to be enhanced, represents the total number of pixels in the j-th sliding window in the image to be enhanced; wherein, the density of Ki-67-positive cells in the j-th sliding window refers to the proportion of the total number of pixels in the fluorescent-labeled area among all pixels in the j-th sliding window.
8. The method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases according to claim 1, characterized in that: The adjustment weight of each pixel point refers to the average of the normalized cell proliferation weights of all sliding windows corresponding to each pixel point.
9. The method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases according to claim 1, characterized in that: The calculation formula of the cropping coefficient of each enhanced area in each image to be enhanced is: Where, represents the cropping coefficient of the u-th enhanced region in the image to be enhanced; Indicates the preset adjustment factor, whose value range is [0.3,0.5]; It represents the mean of the adjustment weights of all pixels in the u-th enhancement area in the image to be enhanced.
10. The method for enhanced recognition of pathological tissue regions based on dynamic epidermal reactions of skin diseases according to claim 1, characterized in that: The specific process of enhancing each image to be enhanced is: taking all enhanced regions in each image to be enhanced and their corresponding cropping coefficients as inputs of the CLAHE algorithm, enhancing all enhanced regions in each image to be enhanced, and obtaining enhanced images to be enhanced.