Pressure damage image detection and evaluation method

Abstract

The invention relates to the technical field of medical image processing and machine learning. The pressure injury image detection and evaluation method comprises the following steps: collecting video image data of the skin of a patient in real time, and preprocessing the collected data to obtain a preprocessed data set; performing video amplification processing on the preprocessed data set to obtain an amplified data set; carrying out image segmentation on the amplified data set, and dividing a plurality of initial pressure damage prediction areas; the mass center position point of each initial damage prediction area is calculated, and the boundary of each initial pressure damage prediction area is dynamically adjusted according to various feature index data of the mass center position point to obtain a pressure damage prediction area; and outputting a detection result corresponding to the pressure injury prediction area in the form of a chart and a report, and giving corresponding early warning information and nursing suggestions according to the detection result. The technical problems that in the prior art, pressure damage image data detection is insufficient in accuracy, and accurate recognition and segmentation are difficult are solved.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing and machine learning technology, and in particular to a pressure injury image detection and evaluation method. Background Art

[0002] Pressure injuries, also known as pressure ulcers or bedsores, are caused by prolonged external pressure, shear force or friction on a part of the body, which results in local damage or necrosis of the skin and subcutaneous tissue in that part. This type of injury is common in patients who are bedridden for a long time, elderly people with limited mobility, and individuals who are under continuous pressure on certain parts of the body. Pressure injuries not only cause great pain to patients, but may also cause infection and other complications, seriously affecting the patient's quality of life.

[0003] Traditional pressure injury detection mainly relies on visual observation and manual evaluation by medical staff, which is not only time-consuming and labor-intensive, but also easily affected by human factors, resulting in insufficient accuracy and reliability of the detection results. With the continuous advancement of medical technology and the rapid development of computer vision technology, people have begun to try to use image processing and machine learning technology to automatically detect pressure injuries.

[0004] However, there are still some problems with existing pressure injury image detection methods. For example, due to the complexity of the patient's skin surface (such as texture, color, etc.) and changes in lighting conditions, the collected video image data often contains a lot of noise and interference information, which increases the difficulty of image processing. In addition, pressure injuries in different patients vary in manifestation, ranging from mild skin redness to severe tissue necrosis, which makes accurate identification and segmentation of the injured area a challenge. Summary of the invention

[0005] The purpose of the present invention is to provide a method for detecting and evaluating pressure injury images, aiming to solve the technical problems in the prior art that pressure injury image data is difficult to process due to the complexity of the patient's skin and changes in lighting conditions, resulting in insufficient detection accuracy, and the pressure injury areas are difficult to accurately identify and segment due to different manifestations.

[0006] The present invention is achieved through the following technical solutions: A method for detecting pressure injury images, comprising the steps of: Collecting video image data of the patient's skin in real time, and preprocessing the collected video image data to obtain a preprocessed data set; Performing video magnification processing on the preprocessed data set to obtain an enlarged data set; The enlarged data set is segmented by clustering algorithm to divide several initial pressure injury prediction areas; Calculate the centroid position point of each initial injury prediction area, and dynamically adjust the boundary of each initial pressure injury prediction area according to various characteristic index data of the centroid position point to obtain the pressure injury prediction area; The test results corresponding to the pressure injury prediction area are output in the form of charts and reports, and corresponding warning information and nursing suggestions are given based on the test results.

[0007] Optionally, the specific process of real-time acquisition of the video image data of the patient's skin and preprocessing the acquired video image data to obtain the preprocessed data set is: The video image data of the patient's skin is collected in real time through a video image sensor equipped with a light source lamp; Preview the collected video image data through the display, and after confirming that the collected content meets the detection requirements, transmit the video image data to the host; The host performs image denoising and brightness equalization processing on the received video image data to obtain a preprocessed data set.

[0008] Optionally, the specific process of performing video magnification processing on the preprocessed data set to obtain the magnified data set is: The video sequence in the preprocessed data set is regarded as a brightness signal in three-dimensional space-time, the brightness signal is filtered in the time domain, and the target frequency component is separated by a bandpass filter; wherein the target frequency component is the characteristic frequency corresponding to the blood flow and tissue movement under the skin in the video sequence; Performing spatial filtering on the luminance signal after time domain filtering; Amplifying the luminance signal after spatial filtering according to a preset amplification factor; The amplified brightness signal is reassembled into video data to form an amplified data set.

[0009] Optionally, the specific process of performing image segmentation on the enlarged data set by using a clustering algorithm to divide a plurality of initial pressure injury prediction areas is: Extract the brightness signal of each frame of the image in the enlarged data set and construct a brightness signal data set; Perform K-means clustering on the brightness signal data set, set the number of clusters to 2, and randomly select two initial cluster centers; Calculate the distance between the brightness signal of each pixel and the two initial cluster centers, and assign each pixel to the initial cluster center closest to it to form two clusters; Update the cluster center by calculating the mean of the brightness signals of all pixels in the cluster; The clustering process is completed by iterative calculation until the cluster center no longer changes significantly or the preset number of iterations is reached; According to the clustering results, the pixels corresponding to the clusters belonging to the potential pressure injury area are divided into the initial pressure injury prediction area.

[0010] Optionally, the specific process of calculating the centroid position point of each initial injury prediction area and dynamically adjusting the boundary of each initial pressure injury prediction area according to various characteristic index data of the centroid position point to obtain the pressure injury prediction area is: Calculate the average values ​​of the horizontal and vertical coordinates of all pixel points in each initial pressure injury prediction area to obtain the corresponding centroid position point; Extract the brightness signal of the centroid point and the corresponding adjacent pixel points, and calculate the time domain statistical characteristics and frequency domain characteristics of the brightness signal to obtain characteristic index data; According to the characteristic index data, the boundary of the initial pressure injury prediction area is adaptively expanded or contracted iteratively adjusted; Through iterative adjustment, the boundaries of all initial pressure injury prediction areas reach a stable state or meet the preset iteration conditions, and finally an accurate pressure injury prediction area is obtained.

[0011] Optionally, the specific process of adaptively expanding or shrinking the boundary of the initial pressure injury prediction area according to the characteristic index data is as follows: According to the relative importance of each feature indicator for pressure injury image detection, a corresponding weighting coefficient is set for each feature indicator; according to the preset weighted fusion rule, the weighted fusion value of the feature indicator is calculated. If the weighted fusion value is greater than the preset expansion threshold, the boundary is expanded; if the weighted fusion value is less than the preset contraction threshold, the boundary is contracted.

[0012] The present invention also provides a pressure injury image assessment method, based on the pressure injury image detection method, comprising the steps of: By using an edge detection algorithm, edge features of the pressure injury prediction area are identified to obtain edge detection results of the pressure injury prediction area; Performing contour recognition on the edge detection results of the pressure injury prediction area, screening out the contour of the pressure injury prediction area that meets the pressure injury characteristics, and obtaining the contour recognition result; Performing texture analysis on the contour of the pressure injury prediction area, extracting the texture features of the contour of the pressure injury prediction area, and analyzing the direct difference between the texture features of the contour of the pressure injury prediction area and the texture features of normal skin; By iteratively calculating the image scaling at different scales and the corresponding gray-level co-occurrence matrix, the texture features of the contour of the pressure injury prediction area are evaluated to obtain the texture feature evaluation results; Combine the edge detection results, contour recognition results and texture feature evaluation results to calculate the misjudgment rate, exclude the misjudgment area according to the preset misjudgment threshold, and obtain the final pressure injury prediction area; The test results corresponding to the final pressure injury prediction area are output in the form of charts, reports or visual interfaces, and the corresponding pressure injury area information, risk level and corresponding nursing recommendations are given.

[0013] Optionally, the specific process of identifying edge features of the pressure injury prediction area by using an edge detection algorithm to obtain an edge detection result of the pressure injury prediction area is: The Canny edge detection algorithm was applied to the pressure injury prediction area. The image in the pressure injury prediction area was smoothed using a Gaussian filter. The amplitude and direction of the image gradient were calculated to identify the intensity changes in the image. The edge was refined using non-maximum suppression, and the edge pixels were determined through double threshold detection. The edge pixels were connected to form a complete edge contour to obtain the edge detection result of the pressure injury prediction area.

[0014] Optionally, the specific process of performing contour recognition on the edge detection result of the pressure injury prediction area to screen out the contour of the pressure injury prediction area that meets the pressure injury characteristics and obtain the contour recognition result is: The contour analysis algorithm is used to perform contour analysis on the edge detection results of the pressure injury prediction area, and the contour feature value of each pressure injury prediction area is calculated; According to the preset shape index of the pressure injury area and the contour analysis results, the shape of the contour of each pressure injury prediction area is evaluated to screen out the contour that meets the typical characteristics of the pressure injury; Calculate the centroid position of the screened contour and compare it with the centroid position of the pressure injury prediction area to verify the accuracy and position consistency of the contour; The contours that meet the shape and position requirements are marked as the pressure injury prediction area contours that meet the pressure injury characteristics to obtain the contour recognition results, and the contours that do not meet the shape or position requirements are regarded as false detections or noise and are excluded.

[0015] Optionally, the specific process of performing texture analysis on the contour of the pressure injury prediction area, extracting texture features of the contour of the pressure injury prediction area, and obtaining the texture feature evaluation result is: Under the selected offset and angle, the co-occurrence frequency of the pixel grayscale values ​​within the contour of the pressure injury prediction area is calculated, and the corresponding grayscale co-occurrence matrix is ​​constructed; According to the corresponding gray-level co-occurrence matrix, the texture feature quantity of the pressure injury prediction area is calculated and analyzed, and compared with the texture feature quantity of normal skin to obtain the corresponding difference value; The image of the pressure injury prediction area is scaled by several scaling ratios, and the gray-level co-occurrence matrix and texture feature quantity corresponding to each scaling ratio are calculated to capture and evaluate the texture features of the pressure injury prediction area at each scaling ratio; The texture feature quantities corresponding to each scaling ratio are compared and analyzed with the preset pressure injury texture feature quantities. According to the matching degree and difference degree of the texture feature quantities, it is evaluated whether there is a risk of pressure injury in the pressure injury prediction area, and the corresponding texture feature evaluation results are given.

[0016] The technical solution of the present invention has at least the following advantages and beneficial effects: Improve detection efficiency and accuracy: By collecting video image data of the patient's skin in real time and preprocessing it, and combining video magnification processing and clustering algorithms for image segmentation, it can automatically identify and divide the predicted area of ​​pressure injury, greatly reducing the workload of medical staff's naked eye observation and manual evaluation, while improving the accuracy and reliability of detection.

[0017] Dynamically adjust the prediction area: By calculating the centroid position point of each initial injury prediction area and dynamically adjusting the boundary of the prediction area according to the characteristic indicator data of the centroid position point, it can better adapt to the complexity of the skin surface of different patients and the diversity of pressure injury manifestations, thereby improving the accuracy of injury area identification.

[0018] Comprehensive assessment and early warning: By performing edge detection, contour recognition and texture analysis on the predicted pressure injury area, the characteristics of the injured area can be comprehensively assessed, and the misjudged area can be effectively eliminated to obtain the final predicted pressure injury area. At the same time, corresponding early warning information and nursing suggestions are given according to the test results, which helps medical staff to take timely intervention measures to reduce patients' pain and the occurrence of complications.

[0019] Visual output and convenience: The test results are output in the form of charts, reports or visual interfaces, providing intuitive and easy-to-understand evaluation results, allowing medical staff to quickly understand the patient's pressure injury and develop personalized care plans based on the evaluation results.

[0020] Strong adaptability: It is not only suitable for long-term bedridden patients and elderly people with limited mobility, but also for other individuals who may suffer from pressure injuries due to continuous pressure on certain parts of the body. It has strong versatility and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of the process of the pressure injury image detection method according to an embodiment of the present invention; Figure 2Schematic diagram of the process of the pressure injury image assessment method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0023] Example 1 Reference Figure 1 , Figure 2 , a pressure injury image detection method, comprising the steps of: Step 1: collect video image data of the patient's skin in real time, and preprocess the collected video image data to obtain a preprocessed data set.

[0024] In this embodiment, the specific process is: The video image data of the patient's skin is collected in real time through a video image sensor equipped with a light source lamp; the role of the light source lamp is to ensure that the collected video image has sufficient brightness and contrast for subsequent processing and analysis. The resolution, frame rate and other parameters of the video image sensor should be selected according to actual needs to ensure that the collected video image has sufficient clarity and details.

[0025] The collected video image data is previewed through the display, and after confirming that the collected content meets the detection requirements, the video image data is transmitted to the host; during the preview process, the angle, focal length and light source brightness of the video image sensor can be adjusted to obtain the best collection effect.

[0026] The host performs image denoising and brightness equalization on the received video image data to obtain a preprocessed data set. The purpose of denoising is to eliminate noise and interference in the video image and improve image quality; brightness equalization makes the brightness distribution in the video image more uniform, which is convenient for subsequent image analysis and processing. Image denoising and brightness equalization algorithms are selected and optimized according to the characteristics of the video image and the needs of subsequent processing.

[0027] Step 2: Perform video magnification processing on the preprocessed data set to obtain an enlarged data set.

[0028] In this embodiment, the specific process is: The video sequence in the preprocessed data set is regarded as a brightness signal in three-dimensional space-time, and the brightness signal is filtered in the time domain, and the target frequency component is separated by a bandpass filter; wherein the target frequency component is the characteristic frequency of the blood flow and tissue movement under the skin in the video sequence; the brightness signal after the time domain filtering is spatially filtered; the brightness signal after the spatial filtering is amplified according to a preset amplification factor; the amplified brightness signal is recombined into video data to form an amplified data set.

[0029] The brightness signal in this embodiment represents the change of the brightness value of each pixel in the video image over time. The brightness signal is filtered in the time domain to separate the characteristic frequencies of blood flow and tissue movement under the skin in the video sequence, which are usually related to the formation and development of pressure injuries. Use a bandpass filter for time domain filtering, and set a suitable bandwidth to capture the target frequency component. The selection of bandwidth is determined according to the characteristic frequency range of pressure injuries. The brightness signal after time domain filtering is spatially filtered to further eliminate noise and interference and improve the signal-to-noise ratio of the signal; spatial filtering can use filters such as Gaussian filters, mean filters, etc., and select appropriate filter types and parameters according to actual needs. The brightness signal after spatial filtering is amplified according to a preset amplification factor. The selection of the amplification factor is determined according to the sensitivity and accuracy requirements of pressure injury detection. The amplification factor of the video can be 50 to 100 times. The amplification process can be to amplify the amplitude of the brightness signal or to amplify the frequency component of the signal, depending on the detection target and the characteristics of the pressure injury. The amplified brightness signal is reassembled into video data to form an amplified data set. This step is to convert the processed signal back to video format for subsequent image segmentation and analysis.

[0030] Step 3: Use clustering algorithm to perform image segmentation on the enlarged data set and divide several initial pressure injury prediction areas.

[0031] In this embodiment, the specific process is: Extract the brightness signal of each frame of the image in the enlarged data set, and take the brightness value of each pixel as a signal element to construct a brightness signal data set; perform K-means clustering on the brightness signal data set, set the number of clusters to 2, and divide the image pixels into two categories, one for normal skin area and the other for potential pressure injury area. Randomly select two initial cluster centers, which can be the brightness values ​​of any two pixels in the image; calculate the distance (such as Euclidean distance) between the brightness signal of each pixel and the two initial cluster centers, and assign each pixel to the nearest initial cluster center to form two clusters; calculate the mean of the brightness signals of all pixels in the cluster, use this mean as the new cluster center, and update the cluster center; repeat the above assignment and update steps, and complete the clustering process by iterative calculation until the cluster center no longer changes significantly (that is, the change in the cluster center is less than a preset threshold) or the preset number of iterations is reached; according to the clustering results, the pixels corresponding to the clusters belonging to the potential pressure injury area are divided into the initial pressure injury prediction area. In this embodiment, the clustering iteration condition may be that the moving distance of the center point is less than 0.0001 or the maximum number of iterations is 300 times, and the maximum number of initialization repetitions is 10 times.

[0032] Step 4: Calculate the centroid position point of each initial injury prediction area, and dynamically adjust the boundary of each initial pressure injury prediction area according to the characteristic index data of the centroid position point to obtain the pressure injury prediction area.

[0033] In this embodiment, the specific process is: Calculate the average values ​​of the horizontal and vertical coordinates of all pixel points in each initial pressure injury prediction area to obtain the corresponding centroid position point; Extract the brightness signals of the centroid point and the corresponding adjacent pixels, that is, extract the brightness signals of the pixels within a certain range around the centroid point. These brightness signals contain important information about blood flow and tissue movement under the skin. Calculate the time domain statistical characteristics and frequency domain characteristics of the brightness signal to obtain characteristic index data; time domain characteristics such as mean, variance, standard deviation, etc.; frequency domain characteristics such as spectrum, power spectrum, etc. Time domain characteristics and frequency domain characteristics can reflect slight changes in skin condition.

[0034] According to the characteristic indicator data, the boundary of the initial pressure injury prediction area is adaptively expanded or contracted iteratively adjusted; in this embodiment, according to the relative importance of each characteristic indicator for the pressure injury image detection, a corresponding weighting coefficient is set for each characteristic indicator, and the weighting coefficient reflects the contribution of different characteristic indicators in assessing the skin condition; according to the preset weighted fusion rule, the weighted fusion value of the characteristic indicator is calculated, and if the weighted fusion value is greater than the preset expansion threshold, the boundary is expanded; if the weighted fusion value is less than the preset contraction threshold, the boundary is contracted. If the weighted fusion value is greater than the expansion threshold, it indicates that there may be a greater risk of pressure injury in the area, so the boundary needs to be expanded to include more possible injury areas. On the contrary, if the weighted fusion value is less than the contraction threshold, it indicates that the risk of the area is low, and the boundary needs to be contracted to reduce the false detection area. The adjustment of the boundary can be performed in an iterative manner, and each adjustment is based on the current weighted fusion value and the threshold comparison result.

[0035] Through iterative adjustment, the boundaries of all initial pressure injury prediction areas reach a stable state or meet the preset iteration conditions (such as the number of iterations, boundary change, etc.). After multiple iterative adjustments, accurate pressure injury prediction areas are finally obtained. These areas more clearly represent the possible pressure injury locations on the image, providing strong support for subsequent analysis and nursing recommendations.

[0036] Step 5: Output the test results corresponding to the pressure injury prediction area in the form of charts and reports, and give corresponding warning information and nursing suggestions based on the test results.

[0037] In this embodiment, the relevant test results are sorted according to the final pressure injury prediction area obtained in the previous steps. According to the actual application scenario and user needs, a suitable output form is selected, which can be a chart (such as a pressure injury distribution map, a risk level map, etc.), a report (a document containing detailed test data and conclusions) or a visual interface (real-time display of test results in software or applications). If a chart or report form is selected, a chart or report is created or written using appropriate graphic design software or report generation tools based on the sorted test results; the chart should clearly show the location, size and shape of the prediction area, as well as information such as the risk level; the report should contain detailed test steps, methods, result analysis and conclusions, and nursing recommendations based on these results. If a visual interface form is selected, a user interface can be designed for real-time display of the test results; the interface includes a real-time image of the prediction area, a risk level indicator, necessary interactive controls (such as zoom in, zoom out, pan, etc.), and a detailed description of the test results, ensuring that the interface design is intuitive and easy to use, so that medical staff can quickly understand and respond to the test results. According to the risk level of the pressure injury prediction area, corresponding warning information is given, and the warning information includes: risk level description, possible injury type, predicted trend of injury development, etc. According to the early warning information, combined with the specific conditions of the patient (such as age, health status, mobility, etc.), targeted nursing advice is given; nursing advice should cover preventive measures (such as adjusting body position, using pressure relief pads, etc.), monitoring frequency, nursing skills, etc. The generated charts, reports or visual interfaces are distributed to relevant medical staff or patients' families to ensure that they can understand these results and suggestions and know how to take corresponding measures based on this information. Follow up the follow-up of the patient, collect feedback from medical staff or patients' families, and continuously optimize the accuracy and practicality of detection methods, output results and suggestions based on the feedback.

[0038] In this embodiment, a pressure injury image assessment method is also provided, based on the above pressure injury image detection method, comprising the steps of: Step 1: Using an edge detection algorithm, identify edge features of the pressure injury prediction area and obtain edge detection results of the pressure injury prediction area.

[0039] In this embodiment, the specific process is: The Canny edge detection algorithm is applied to the pressure injury prediction area; the image in the pressure injury prediction area is smoothed using a Gaussian filter. The purpose of the Gaussian filter is to reduce the impact of noise on edge detection while retaining important edge features of the image. The magnitude and direction of the image gradient are calculated to identify the intensity change in the image; the gradient calculation can be achieved through the Sobel operator, Prewitt operator or Roberts operator, etc. The gradient magnitude indicates the speed of the intensity change in the image, while the gradient direction indicates the direction of the intensity change. The edge is refined using non-maximum suppression, and only the maximum value point in the gradient direction is retained, thereby removing redundant edge points; by comparing the gradient magnitude of each pixel point with its adjacent pixel points in the gradient direction, the pixel point with the largest gradient magnitude is retained. The edge pixels are determined by dual threshold detection, and the edge pixels are connected to form a complete edge contour to obtain the edge detection result of the pressure injury prediction area. Two thresholds can be set: a high threshold and a low threshold; first, a high threshold is used for preliminary edge detection to obtain a set of strong edge points; then, a low threshold is used for complementary edge point detection to obtain a set of weak edge points; among the weak edge points, a complete edge contour is formed by connecting the points adjacent to the strong edge points. After double threshold detection, a series of discrete edge pixel points are obtained, and these discrete edge pixel points are connected using edge connection algorithms (such as Hough transform, contour tracking, etc.) to form a complete edge contour. The obtained edge contour is output as the edge detection result of the pressure injury prediction area; the edge detection result can be displayed in the form of an image or stored in the form of a coordinate set of edge points.

[0040] Step 2: Perform contour recognition on the edge detection results of the pressure injury prediction area, screen out the contour of the pressure injury prediction area that meets the pressure injury characteristics, and obtain the contour recognition result.

[0041] In this embodiment, the specific process is: The edge detection results of the pressure injury prediction area are subjected to contour analysis through the contour analysis algorithm, and the contour feature values ​​of each pressure injury prediction area are calculated; the contour feature values ​​include area, perimeter, shape factor (such as the ratio of area to perimeter), roundness (such as the ratio of the contour perimeter to the perimeter of a circle of the same area), etc. These contour features can reflect the basic form and complexity of the contour, and provide a quantitative basis for subsequent screening.

[0042] According to the preset shape index of the pressure injury area and the contour analysis results, the shape of each pressure injury prediction area contour is evaluated to screen out contours that meet the typical characteristics of pressure injuries. Among them, the preset shape index of the pressure injury area can be obtained through clinical experience or literature research; shape indexes include area range, shape factor threshold, roundness threshold, etc.

[0043] The centroid position of the screened contours was calculated and compared with the centroid position of the pressure injury prediction area to verify the accuracy and position consistency of the contours and further eliminate falsely detected contours caused by image deformation or processing errors.

[0044] Contours that meet the shape and position requirements are marked as pressure injury prediction area contours that meet the pressure injury characteristics to obtain contour recognition results, and contours that do not meet the shape or position requirements are regarded as false detections or noise and excluded; the reasons for false detections are recorded for subsequent optimization of the algorithm or adjustment of detection parameters.

[0045] Step 3: Perform texture analysis on the contour of the pressure injury prediction area, extract the texture features of the contour of the pressure injury prediction area, and obtain the texture feature evaluation result.

[0046] In this embodiment, the specific process of performing texture analysis on the contour of the pressure injury prediction area is as follows: Under the selected offset and angle, the co-occurrence frequency of the pixel grayscale values ​​within the outline of the pressure injury prediction area is calculated, and the corresponding grayscale co-occurrence matrix is ​​constructed. According to the characteristics of the pressure injury image and the analysis requirements, the appropriate grayscale level (such as 256 grayscale levels), offset (such as the distance between pixels) and angle (such as 0°, 45°, 90°, 135°, etc.) are selected; each pixel within the outline of the pressure injury prediction area is traversed, and the frequency of its grayscale value combination with the adjacent pixels under the selected offset and angle is calculated to construct the grayscale co-occurrence matrix, which reflects the spatial distribution and texture characteristics of the grayscale values ​​in the image.

[0047] According to the corresponding gray-level co-occurrence matrix, the texture feature quantities of the pressure injury prediction area are calculated and analyzed, and compared with the texture feature quantities of normal skin to obtain the corresponding difference values; texture feature quantities such as contrast, correlation, energy (or homogeneity), entropy, etc., can quantitatively describe the texture characteristics of the image. According to the characteristics of the pressure injury image and the detection requirements, the texture feature quantities that are sensitive to pressure injuries are selected as the basis for subsequent analysis. Establish or obtain a database of texture feature quantities of normal skin as a comparison benchmark; compare the texture feature quantities of the pressure injury prediction area with the texture feature quantities of normal skin, and calculate the difference value or similarity index.

[0048] The image of the pressure injury prediction area is scaled using several scaling ratios, and the gray-level co-occurrence matrix and texture feature quantity corresponding to each scaling ratio are calculated to capture and evaluate the texture features of the pressure injury prediction area at each scaling ratio.

[0049] The texture feature quantity corresponding to each scaling ratio is compared and analyzed with the preset pressure injury texture feature quantity. According to the matching degree and difference degree of the texture feature quantity, the risk of pressure injury in the pressure injury prediction area is evaluated, and the corresponding texture feature evaluation result is given. The threshold of texture feature difference or similarity can be set to determine whether the pressure injury prediction area is a pressure injury.

[0050] Step 4: Combine the edge detection results, contour recognition results and texture feature evaluation results to calculate the misjudgment rate, exclude the misjudgment area according to the preset misjudgment threshold, and obtain the final pressure injury prediction area.

[0051] In this embodiment, the edge detection results, contour recognition results and texture feature evaluation results are integrated into a unified evaluation framework. Data from different sources (such as the coordinates of edge contours, the classification labels of contour recognition, the difference values ​​of texture features, etc.) are aligned and matched to ensure that they correspond to the same pressure injury prediction area. An indicator of misjudgment rate is defined to quantify the errors that may occur in the edge detection, contour recognition and texture analysis steps. The misjudgment rate can be calculated based on a variety of factors, such as the continuity of the edge contour, the degree of matching between the contour and the preset shape index, the difference value between the texture feature and the normal skin feature, etc. The misjudgment rate of each step is calculated by comparing the actual detection results with the known real situation (such as the pressure injury area obtained by expert annotation or clinical verification). If the real situation is unknown, a cross-validation method can be used to divide the data set into a training set and a test set, use the training set for model training, and then calculate the misjudgment rate on the test set. One or more misjudgment thresholds are set according to clinical needs and detection accuracy requirements. These thresholds are used to determine which detection results may be misjudged and exclude these areas accordingly. The setting of the false positive threshold should take into account multiple factors, including the sensitivity, specificity, accuracy of the detection method and the acceptable range for clinical application. For each pressure injury prediction area, the comprehensive false positive rate is calculated based on its edge detection results, contour recognition results and texture feature evaluation results. If the comprehensive false positive rate of a certain area exceeds the preset false positive threshold, it is regarded as a false positive area and excluded from the final prediction results. After excluding the false positive areas, the final pressure injury prediction areas are obtained, which have high detection accuracy and clinical application value.

[0052] Step 5: Output the test results corresponding to the final pressure injury prediction area in the form of a chart, report or visual interface, and provide the corresponding pressure injury area information, risk level and corresponding nursing recommendations.

[0053] In this embodiment, the data of the final pressure injury prediction area is sorted, including key information such as the coordinates, size, shape characteristics, and texture characteristics of the prediction area. Based on the key information, the risk level of pressure injury is calculated. The risk level can be divided based on the degree of matching between the characteristics of the prediction area and the known pressure injury characteristics, parameters such as the area and perimeter of the prediction area, and possible lesion development trends. Based on the sorted data, intuitive charts are generated, such as distribution maps of pressure injury prediction areas, heat maps of risk levels, etc., so that medical staff can quickly understand the patient's skin condition. Write a detailed test report, including the patient's basic information, test time, test process, specific information of the prediction area (such as location, size, shape, texture characteristics, etc.), risk level assessment, and possible nursing suggestions. According to the risk level of pressure injury and the specific characteristics of the prediction area, formulate personalized nursing suggestions. The suggestions should cover preventive measures (such as regular turning over, use of pressure relief pads, etc.), monitoring frequency, treatment plans (such as drug therapy, physical therapy, etc.), etc.

[0054] Example 2 Based on Example 1, in this embodiment, the specific steps of performing image segmentation on the enlarged data set by using a clustering algorithm are as follows: Extract the brightness signal of each frame of the image in the enlarged data set and construct a brightness signal data set; suppose that in the extracted enlarged data set, the target area in the first frame of the video stream contains pixels, the brightness signal of each pixel can be regarded as a one-dimensional data, then The set of brightness values ​​of pixels can be expressed as ,but Represents the brightness value of the pixel, ,and .

[0055] Perform K-means clustering on the brightness signal data set, set the number of clusters to 2, and randomly select two initial cluster centers, denoted as and , ; Calculate the distance between the brightness signal of each pixel and the two initial cluster centers, and assign each pixel to the initial cluster center with the closest distance to form two clusters, cluster 1 and cluster 2. The distance can be calculated using the Euclidean distance, as shown in the following formula (1):

[0056] in, Represents pixel With cluster center The distance value between Represents a pixel in an image, with its brightness value as one-dimensional data; Represents the cluster center, which is also a one-dimensional data, representing the mean brightness value of all pixels in the cluster; Represents pixel With cluster center The absolute value of the difference between them, that is, the distance between them, here, due to and They are all one-dimensional data (brightness value), so the distance calculation is simplified to the absolute value of the difference.

[0057] like , then the pixel Assign to cluster 1, otherwise assign to cluster 2; update the cluster center by calculating the mean of the brightness signal of all pixels in the cluster. For each cluster, recalculate its new cluster center, that is, the mean of all data points in the cluster. Assume that the point in cluster 1 is , the points in cluster 2 are , then the expression of the new cluster center is as shown in the following formula (2) and formula (3):

[0058]

[0059] in, and are the new cluster centers of cluster 1 and cluster 2 respectively. and are the number of data points in cluster 1 and cluster 2 respectively; The clustering process is completed through iterative calculation until the cluster center no longer changes significantly or the preset number of iterations is reached; according to the clustering results, the pixel points corresponding to the clusters belonging to the potential pressure injury area are divided into the initial pressure injury prediction area.

[0060] In this embodiment, the centroid position point of each initial injury prediction area is calculated, and the boundary of each initial pressure injury prediction area is dynamically adjusted according to the characteristic index data of the centroid position point. The specific process of obtaining the pressure injury prediction area is as follows: The average values ​​of the horizontal and vertical coordinates of all pixels in each initial pressure injury prediction area are calculated to obtain the corresponding centroid position point. The expression of the centroid position point is shown in the following formula (4):

[0061] in, The horizontal coordinate of the centroid position point; The ordinate of the point representing the centroid position; Indicates the image The horizontal coordinate of the pixel point; Indicates the image The vertical coordinate of the pixel; Represents the total number of pixels in the initial pressure injury prediction area segmented by clustering.

[0062] The brightness signals of the centroid position point and the corresponding adjacent pixel points are extracted, and the time domain statistical characteristics and frequency domain characteristics of the brightness signal are calculated to obtain the characteristic index data.

[0063] According to the characteristic index data, the boundary of the initial pressure injury prediction area is adaptively expanded or contracted and iteratively adjusted; the expression for adaptive expansion or contraction of the boundary is shown in the following formula (5):

[0064] in, Represents the weighted fusion value of the feature index. If If it is greater than the threshold value 1.2, the boundary is expanded. If it is less than 0.8, the boundary shrinks; represents the weighting coefficient; It represents the characteristic index selected from the time domain index and the frequency domain. Through iterative adjustment, until the boundaries of all initial pressure injury prediction areas reach a stable state or meet the preset iteration conditions, an accurate pressure injury prediction area is finally obtained.

[0065] In this embodiment, the edge features of the pressure injury prediction area are identified by the edge detection algorithm, and the specific process of obtaining the edge detection result of the pressure injury prediction area is as follows: The Canny edge detection algorithm is applied to the pressure injury prediction area; the image in the pressure injury prediction area is smoothed using a Gaussian filter; the Gaussian filter is set to , the original image is , then the smoothed image expression is as follows (6):

[0066] in, Represents the smoothed image; calculates the magnitude and direction of the image gradient and identifies the intensity change in the image; assumes that the horizontal and vertical gradients of the image are and , then the expressions of the magnitude and direction of the image gradient are shown in the following equations (7) and (8):

[0067]

[0068] in, Represents the magnitude of the image gradient; and Respectively represent the image in horizontal ( direction) and vertical ( direction); Represents the direction of the image gradient.

[0069] The edge was refined using non-maximum suppression, and the edge pixels were determined through double threshold detection. The edge pixels were connected to form a complete edge contour, and the edge detection result of the pressure injury prediction area was obtained.

[0070] In this embodiment, contour recognition is performed on the edge detection result of the pressure injury prediction area to screen out the contour of the pressure injury prediction area that meets the pressure injury characteristics. The specific process of obtaining the contour recognition result is as follows: The contour analysis algorithm is used to perform contour analysis on the edge detection results of the pressure injury prediction area, and the contour feature value of each pressure injury prediction area is calculated. The expected area and shape expressions of the pressure injury area are shown in the following equations (9) and (10):

[0071]

[0072] in, represents the expected area of ​​the pressure injury area; Indicates The area of ​​the contour of the predicted pressure injury area; Indicates The outline of the identified predicted pressure injury area; represents the desired shape of the pressure injury area; Indicates The circumference of the contour of the identified pressure injury prediction area; according to the expected area and shape of the pressure injury area, the roundness expression is defined as shown in the following formula (11):

[0073] in, Indicates roundness. According to the preset shape index of the pressure injury area and the contour analysis results, the shape of each pressure injury prediction area contour is evaluated to screen out contours that meet the typical characteristics of pressure injury; the centroid position of the screened contour is calculated and compared with the centroid position of the pressure injury prediction area to verify the accuracy and position consistency of the contour; the contours that meet the shape and position requirements are marked as the pressure injury prediction area contours that meet the pressure injury characteristics to obtain the contour recognition results, and the contours that do not meet the shape or position requirements are regarded as false detections or noise and excluded.

[0074] In this embodiment, the specific process of performing texture analysis on the contour of the pressure injury prediction area, extracting the texture features of the contour of the pressure injury prediction area, and obtaining the texture feature evaluation result is as follows: Under the selected offset and angle, the co-occurrence frequency of the pixel grayscale values ​​within the contour of the pressure injury prediction area is calculated, and the corresponding grayscale co-occurrence matrix is ​​constructed; according to the corresponding grayscale co-occurrence matrix, the texture feature of the pressure injury prediction area is calculated and analyzed, and compared with the texture feature of normal skin to obtain the corresponding difference value; the grayscale co-occurrence matrix is ​​calculated under the preset offset and angle, and the grayscale values ​​of the pixel pairs are set to be and , gray-level co-occurrence matrix Grayscale and The co-occurrence frequency of gray level co-occurrence matrix includes contrast, similarity, energy, entropy, etc. The calculation expression of contrast is shown in the following formula (12):

[0075] in, Indicates the contrast value; represents two gray levels in the gray-level co-occurrence matrix; The calculation expression of similarity is shown in the following formula (13):

[0076] in, Represents the similarity value; the similarity value reflects the frequency of occurrence of pixel pairs with similar grayscale values ​​in the image, that is, the uniformity or consistency of the image texture. The higher the similarity value, the more pixels with similar grayscale values ​​in the image and the more uniform the texture. The calculation expression of energy is shown in the following formula (14):

[0077] in, Indicates the energy value; the energy value reflects the uniformity of the grayscale distribution of the image and the coarseness of the texture. The higher the energy value, the more uniform the grayscale distribution in the image and the finer the texture. The calculation expression of entropy is shown in the following formula (15):

[0078] in, Represents the entropy value; the entropy value reflects the complexity of the grayscale distribution of the image and the non-uniformity of the texture. The higher the entropy value, the more complex the grayscale distribution in the image and the more non-uniform the texture.

[0079] The image of the pressure injury prediction area is scaled using several scaling ratios, and the gray-level co-occurrence matrix and texture feature quantity corresponding to each scaling ratio are calculated to capture and evaluate the texture features of the pressure injury prediction area at each scaling ratio; the texture feature quantity corresponding to each scaling ratio is compared and analyzed with the preset pressure injury texture feature quantity, and based on the matching degree and difference degree of the texture feature quantity, it is evaluated whether there is a risk of pressure injury in the pressure injury prediction area, and the corresponding texture feature evaluation results are given.

[0080] Assume that the edge detection result of each pressure injury prediction area is , the contour recognition result is , the texture analysis results are , define the false positive rate The comprehensive evaluation expression of is shown in formula (16):

[0081] in, Indicates the number of feature points; Indicates The weight of the edge detection result of feature points; Indicates The edge detection results of feature points; Indicates The weight of the contour recognition result of feature points; Indicates The contour recognition result of feature points; Indicates The weight of the texture analysis result of each feature point; Indicates Texture analysis results of feature points; satisfy The weight distribution can be adjusted according to the specific application scenario and algorithm performance. For example, if the performance of the edge detection algorithm is particularly good, it can be given a higher weight; conversely, if the performance of the texture analysis algorithm is better, it can also be given a higher weight. , and It can be calculated by comparing the actual test results with the known real situation (such as the pressure injury area obtained through expert annotation or clinical verification). If the real situation is unknown, the cross-validation method can be used to divide the data set into a training set and a test set, use the training set for model training, and then calculate the misjudgment rate on the test set. When the misjudgment rate exceeds the preset misjudgment threshold, the corresponding pressure injury prediction area is excluded, otherwise it is retained. Through multi-level feature analysis, shape and texture evaluation, the misjudgment area in pressure injury detection can be effectively reduced.

Claims

1. A method for detecting pressure injury images, characterized in that: Includes steps: Collecting video image data of the patient's skin in real time, and preprocessing the collected video image data to obtain a preprocessed data set; Performing video magnification processing on the preprocessed data set to obtain an enlarged data set; The enlarged data set is segmented by clustering algorithm to divide several initial pressure injury prediction areas; Calculate the centroid position point of each initial injury prediction area, and dynamically adjust the boundary of each initial pressure injury prediction area according to various characteristic index data of the centroid position point to obtain the pressure injury prediction area; The test results corresponding to the pressure injury prediction area are output in the form of charts and reports, and corresponding warning information and nursing suggestions are given based on the test results.

2. The pressure injury image detection method according to claim 1, characterized in that: The specific process of collecting the video image data of the patient's skin in real time and preprocessing the collected video image data to obtain the preprocessed data set is as follows: The video image data of the patient's skin is collected in real time through a video image sensor equipped with a light source lamp; Preview the collected video image data through the display, and after confirming that the collected content meets the detection requirements, transmit the video image data to the host; The host performs image denoising and brightness equalization processing on the received video image data to obtain a preprocessed data set.

3. The pressure injury image detection method according to claim 1, characterized in that: The specific process of performing video amplification processing on the preprocessed data set to obtain the amplified data set is as follows: The video sequence in the preprocessed data set is regarded as a brightness signal in three-dimensional space-time, the brightness signal is filtered in the time domain, and the target frequency component is separated by a bandpass filter; wherein the target frequency component is the characteristic frequency corresponding to the blood flow and tissue movement under the skin in the video sequence; Performing spatial filtering on the luminance signal after time domain filtering; Amplifying the luminance signal after spatial filtering according to a preset amplification factor; The amplified brightness signal is reassembled into video data to form an amplified data set.

4. The pressure injury image detection method according to claim 1, characterized in that: The specific process of performing image segmentation on the enlarged data set by using a clustering algorithm to divide a number of initial pressure injury prediction areas is as follows: Extract the brightness signal of each frame of the image in the enlarged data set and construct a brightness signal data set; Perform K-means clustering on the brightness signal data set, set the number of clusters to 2, and randomly select two initial cluster centers; Calculate the distance between the brightness signal of each pixel and the two initial cluster centers, and assign each pixel to the initial cluster center closest to it to form two clusters; Update the cluster center by calculating the mean of the brightness signals of all pixels in the cluster; The clustering process is completed by iterative calculation until the cluster center no longer changes significantly or the preset number of iterations is reached; According to the clustering results, the pixels corresponding to the clusters belonging to the potential pressure injury area are divided into the initial pressure injury prediction area.

5. The pressure injury image detection method according to claim 1, characterized in that: The specific process of calculating the centroid position point of each initial injury prediction area and dynamically adjusting the boundary of each initial pressure injury prediction area according to the characteristic index data of the centroid position point to obtain the pressure injury prediction area is as follows: Calculate the average values ​​of the horizontal and vertical coordinates of all pixel points in each initial pressure injury prediction area to obtain the corresponding centroid position point; Extract the brightness signal of the centroid point and the corresponding adjacent pixel points, and calculate the time domain statistical characteristics and frequency domain characteristics of the brightness signal to obtain characteristic index data; According to the characteristic index data, the boundary of the initial pressure injury prediction area is adaptively expanded or contracted iteratively adjusted; Through iterative adjustment, the boundaries of all initial pressure injury prediction areas reach a stable state or meet the preset iteration conditions, and finally an accurate pressure injury prediction area is obtained.

6. The pressure injury image detection method according to claim 5, characterized in that: The specific process of adaptively expanding or shrinking the boundary of the initial pressure injury prediction area according to the characteristic index data is as follows: According to the relative importance of each feature indicator for pressure injury image detection, a corresponding weighting coefficient is set for each feature indicator; according to the preset weighted fusion rule, the weighted fusion value of the feature indicator is calculated. If the weighted fusion value is greater than the preset expansion threshold, the boundary is expanded; if the weighted fusion value is less than the preset contraction threshold, the boundary is contracted.

7. A pressure injury image assessment method, based on the pressure injury image detection method according to any one of claims 1 to 6, characterized in that: Includes steps: By using an edge detection algorithm, edge features of the pressure injury prediction area are identified to obtain edge detection results of the pressure injury prediction area; Performing contour recognition on the edge detection results of the pressure injury prediction area, screening out the contour of the pressure injury prediction area that meets the pressure injury characteristics, and obtaining the contour recognition result; Performing texture analysis on the contour of the pressure injury prediction area, extracting the texture features of the contour of the pressure injury prediction area, and obtaining a texture feature evaluation result; Combine the edge detection results, contour recognition results and texture feature evaluation results to calculate the misjudgment rate, exclude the misjudgment area according to the preset misjudgment threshold, and obtain the final pressure injury prediction area; The test results corresponding to the final pressure injury prediction area are output in the form of charts, reports or visual interfaces, and the corresponding pressure injury area information, risk level and corresponding nursing recommendations are given.

8. The pressure injury image assessment method according to claim 7, characterized in that: The specific process of identifying the edge features of the pressure injury prediction area by the edge detection algorithm and obtaining the edge detection result of the pressure injury prediction area is as follows: The Canny edge detection algorithm was applied to the pressure injury prediction area. The image in the pressure injury prediction area was smoothed using a Gaussian filter. The amplitude and direction of the image gradient were calculated to identify the intensity changes in the image. The edge was refined using non-maximum suppression, and the edge pixels were determined through double threshold detection. The edge pixels were connected to form a complete edge contour to obtain the edge detection result of the pressure injury prediction area.

9. The pressure injury image assessment method according to claim 7, characterized in that: The specific process of performing contour recognition on the edge detection result of the pressure injury prediction area, screening out the contour of the pressure injury prediction area that meets the pressure injury characteristics, and obtaining the contour recognition result is as follows: The contour analysis algorithm is used to perform contour analysis on the edge detection results of the pressure injury prediction area, and the contour feature value of each pressure injury prediction area is calculated; According to the preset shape index of the pressure injury area and the contour analysis results, the shape of the contour of each pressure injury prediction area is evaluated to screen out the contour that meets the typical characteristics of the pressure injury; Calculate the centroid position of the screened contour and compare it with the centroid position of the pressure injury prediction area to verify the accuracy and position consistency of the contour; The contours that meet the shape and position requirements are marked as the pressure injury prediction area contours that meet the pressure injury characteristics to obtain the contour recognition results, and the contours that do not meet the shape or position requirements are regarded as false detections or noise and are excluded.

10. The pressure injury image assessment method according to claim 7, characterized in that: The specific process of performing texture analysis on the contour of the pressure injury prediction area, extracting the texture features of the contour of the pressure injury prediction area, and obtaining the texture feature evaluation result is as follows: Under the selected offset and angle, the co-occurrence frequency of the pixel grayscale values ​​within the contour of the pressure injury prediction area is calculated, and the corresponding grayscale co-occurrence matrix is ​​constructed; According to the corresponding gray-level co-occurrence matrix, the texture feature quantity of the pressure injury prediction area is calculated and analyzed, and compared with the texture feature quantity of normal skin to obtain the corresponding difference value; The image of the pressure injury prediction area is scaled by several scaling ratios, and the gray-level co-occurrence matrix and texture feature quantity corresponding to each scaling ratio are calculated to capture and evaluate the texture features of the pressure injury prediction area at each scaling ratio; The texture feature quantities corresponding to each scaling ratio are compared and analyzed with the preset pressure injury texture feature quantities. According to the matching degree and difference degree of the texture feature quantities, it is evaluated whether there is a risk of pressure injury in the pressure injury prediction area, and the corresponding texture feature evaluation results are given.

Images