Hyperspectral image-based internal fistula failure identification method and system

Abstract

The present application relates to the technical field of image processing, in particular to a method and system for identifying internal fistula failure based on hyperspectral images. The method comprises the following steps: obtaining spectral image data and heart rate curve data of a target region from a preset database; preprocessing the spectral image data and dividing the heart rate curve data into heartbeat state wave bands; selecting spectral images in the same heartbeat state wave band from the spectral image data to form a same-state image set; extracting the hotspot distribution features of each spectral image in the same-state image set to obtain corresponding hotspot distribution maps and determine the abnormal risk evaluation values of the pixel regions in each spectral image; forming an abnormal risk distribution feature map of the abnormal risk evaluation values higher than a preset abnormal risk threshold value and inputting the abnormal risk distribution feature map into a pre-trained image classification model to obtain the image feature type probability distribution of the target region. The present application improves the accuracy of extracting key features for failure risk identification.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to a method and system for identifying arteriovenous fistula malfunction based on hyperspectral images. Background Technology

[0002] Arteriovenous fistula (AVF) failure typically refers to the loss of function of an AVF due to thrombosis, stenosis, or occlusion, making effective hemodialysis impossible. Hyperspectral imaging (HSI) is a device capable of capturing broadband spectral data, effectively monitoring the absorption and reflection peaks of substances at specific wavelengths to obtain their "spectral fingerprints." Since HSI can distinguish the spectral characteristics of oxyhemoglobin and deoxyhemoglobin, and because blood flow obstruction and changes in local oxygenation occur in the early stages of dialysis-related AVF failure, HSI has significant potential for identifying and providing early warning of AVF failure.

[0003] Traditionally, the identification of arteriovenous fistula (AVF) failure mainly relies on Doppler ultrasound to measure parameters such as blood flow velocity and volume to determine whether failure has occurred. This requires applying coupling gel to the skin surface of the AVF and applying pressure to the ultrasound probe. Frequent high-frequency failure testing may affect blood flow in the fragile fistula and cause additional complications, especially posing a higher risk to patients with existing thrombosis or aneurysm. Furthermore, operating the ultrasound probe requires professional personnel, and the calculation of angles and sampling volumes depends on the operator's experience, resulting in high errors in the calculation of failure parameters and making it difficult to standardize the test results. Summary of the Invention

[0004] To address the technical problem of low accuracy in identifying arteriovenous fistula (AVF) failure by extracting key features, the present invention aims to provide a method and system for identifying AVF failure based on hyperspectral images. The specific technical solution adopted is as follows: In a first aspect, embodiments of the present invention provide a method for identifying arteriovenous fistula failure based on hyperspectral images, the method comprising: Spectral image data of the target area is obtained from a preset database, and heart rate curve data synchronized with the spectral image acquisition period is obtained simultaneously. The spectral image data is preprocessed, and the heart rate curve data is divided into heartbeat state bands. Based on the divided heartbeat state bands, spectral images in the same heartbeat state band are selected from the spectral image data to form a set of images in the same state. Extract the hotspot distribution features of each spectral image in the same state image set to obtain the hotspot distribution map corresponding to each spectral image, and determine the abnormal risk assessment value of the pixel region in each spectral image based on the difference between the hotspot distribution maps corresponding to different spectral images. An anomaly risk assessment value that exceeds a preset anomaly risk threshold is used to form an anomaly risk distribution feature map. The anomaly risk distribution feature map is then input into a pre-trained image classification model to obtain the probability distribution of image feature types in the target region.

[0005] In some embodiments, acquiring spectral image data of the target region from a preset database and simultaneously acquiring heart rate curve data synchronized with the spectral image acquisition period includes: Retrieve spectral image data of the target region from a preset database; Retrieve heart rate curve data that was synchronously acquired with the spectral image data from the preset database; Establish the correspondence between each frame of spectral image data and the time points of heart rate curve data; Verify the time synchronization accuracy between spectral image data and heart rate curve data; Store and manage the time-synchronized spectral image data and heart rate curve data.

[0006] In some embodiments, the preprocessing of the spectral image data includes: Image noise reduction processing is performed on the acquired spectral image data of the target area; Perform grayscale correction on the spectral image data of the target area after noise reduction; Spatial registration processing is performed on the spectral image data of the target region after grayscale correction; The spectral image data after image denoising, grayscale correction and spatial registration is used as the final spectral image data.

[0007] In some embodiments, the step of dividing the heart rate curve data into heart rate state bands includes: A one-dimensional CNN network is constructed, and the heart rate curve data is processed using the one-dimensional CNN network to identify the starting index of each wave segment of the heart rate curve. The heart rate curve data is divided into heart rate state bands based on the starting index.

[0008] In some embodiments, the step of selecting spectral images belonging to the same heartbeat state band from the spectral image data based on the divided heartbeat state bands to form a set of images with the same heartbeat state includes: Establish the correlation between the heartbeat state bands and the time of spectral image data acquisition; Based on the aforementioned correlation, all spectral images acquired at the same heartbeat state band are selected. The spectral images that are in the same heartbeat state band are grouped together. Each group of spectral images belonging to the same heartbeat state band is identified to clarify the heartbeat state band to which the spectral image belongs; Complete the screening and grouping of spectral images corresponding to all heartbeat state bands to form multiple sets of images in the same state.

[0009] In some embodiments, extracting the hotspot distribution features of each spectral image in the same state image set to obtain a hotspot distribution map corresponding to each spectral image includes: Determine the preset feature bands used to extract hotspot distribution characteristics; The spectral images in the same state image set are analyzed based on the preset feature bands; For each pixel in the spectral image, a neighborhood growing analysis is performed, and hotspot pixels are determined based on the difference between the pixel's gray value and the gray values ​​of its neighboring pixels. All identified hotspot pixels are combined to form a hotspot distribution map corresponding to the spectral image.

[0010] In some embodiments, determining the anomaly risk assessment value of pixel regions in each spectral image based on the hotspot distribution map includes: Hotspot pixels and non-hotspot pixels are marked according to the hotspot distribution map; The degree of abnormal risk corresponding to the hotspot distribution map is calculated based on the number of hotspot pixels and the number of non-hotspot pixels. Based on the aforementioned level of anomaly risk, an anomaly risk assessment value is determined for each pixel region in each spectral image.

[0011] In some embodiments, forming an abnormal risk distribution feature map from abnormal risk assessment values ​​that exceed a preset abnormal risk threshold includes: From the abnormal risk assessment values ​​corresponding to the spectral image, abnormal risk assessment values ​​that are higher than the preset abnormal risk threshold are selected and mapped to the original pixel region position of the spectral image. An initial abnormal risk distribution feature map is constructed based on the abnormal risk assessment value that exceeds the preset abnormal risk threshold and its corresponding pixel region location; The initial abnormal risk distribution feature map is subjected to invalid region removal processing, and the processed abnormal risk distribution feature map is used as the final abnormal risk distribution feature map.

[0012] In some embodiments, inputting the abnormal risk distribution feature map into a pre-trained image classification model to obtain the image feature type probability distribution of the target region includes: The abnormal risk distribution feature map is input into a pre-trained image classification model; The image classification model is used to extract and analyze the features of the abnormal risk distribution feature map, and the features of the abnormal risk distribution feature map are compared with those of historical samples in a preset historical data sample library. The image classification model calls a preset multi-patient cross-sectional data sample library and compares the input feature map with the cross-sectional sample features. By combining the comparison results of historical samples and cross-sectional samples, the pre-trained image classification model calculates the risk probability of each type of arteriovenous fistula failure. Output the risk probability of each type of failure corresponding to the arteriovenous fistula region to form the probability distribution of image feature types of the target region.

[0013] Secondly, embodiments of the present invention provide a fistula failure identification system based on hyperspectral images, the system comprising the following modules: The acquisition module is used to acquire spectral image data of the target area from a preset database, and simultaneously acquire heart rate curve data synchronized with the spectral image acquisition period; The segmentation module is used to preprocess the spectral image data and segment the heart rate curve data into heartbeat state bands. The filtering module is used to filter out spectral images in the same heartbeat state band from the spectral image data according to the divided heartbeat state bands, and form a set of images in the same state. The extraction module is used to extract the hotspot distribution features of each spectral image in the same state image set, obtain the hotspot distribution map corresponding to each spectral image, and determine the abnormal risk assessment value of the pixel region in each spectral image based on the difference between the hotspot distribution maps corresponding to different spectral images. The generation module is used to form an abnormal risk distribution feature map from abnormal risk assessment values ​​that are higher than a preset abnormal risk threshold, and input the abnormal risk distribution feature map into a pre-trained image classification model to obtain the probability distribution of image feature types of the target region.

[0014] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements the various possible implementations of the first aspect.

[0015] Fourthly, embodiments of the present invention provide a computer program product comprising: computer program code, which, when executed on a computer, causes the computer to perform the method described in the first aspect or any possible implementation thereof.

[0016] Fifthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the various possible implementations of the first aspect.

[0017] The embodiments of the present invention have at least the following beneficial effects: This invention acquires synchronized arteriovenous fistula (AVF) spectral image data and heart rate curve data from a pre-set database. First, preprocessing eliminates spectral image noise, grayscale deviation, and spatial offset. Then, a one-dimensional CNN network is used to divide the heart rate curve into heartbeat state bands, ensuring subsequent analysis is based on a consistent physiological state. Next, spectral images with the same heartbeat state bands are selected to form a set, avoiding interference from heart rate fluctuations. Subsequently, hotspot distribution features are extracted, and the difference is calculated using the DTW algorithm to accurately determine the abnormal risk assessment value of pixel regions. Finally, an abnormal risk distribution feature map is formed through threshold filtering, input into a pre-trained multi-layer CNN model, and the output failure type probability distribution is obtained. The entire process involves contactless acquisition and focuses on key features, avoiding damage to the fistula caused by contact detection and reducing human error, significantly improving the accuracy and standardization of AVF failure identification. Attached Figure Description

[0018] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a method for identifying arteriovenous fistula failure based on hyperspectral images, provided in an embodiment of the present invention. Figure 2 This is a system block diagram of an arteriovenous fistula failure identification system based on hyperspectral images, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of a computer device provided in one embodiment of the present invention. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the arteriovenous fistula failure identification method and system based on hyperspectral images proposed in this invention.

[0021] In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form.

[0022] In the description of the embodiments of the present invention, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of the present invention, "multiple" means two or more.

[0023] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0024] Unless otherwise defined, 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 invention pertains.

[0025] The embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided by the embodiments of the present invention are also applicable to similar technical problems.

[0026] The following description, in conjunction with the accompanying drawings, details the specific scheme of the arteriovenous fistula failure identification method and system provided by this invention.

[0027] Example 1: Please see Figure 1 The diagram illustrates a flowchart of a method for identifying arteriovenous fistula failure based on hyperspectral images according to an embodiment of the present invention. The method includes the following steps: This invention provides a method for identifying arteriovenous fistula failure based on hyperspectral images, the method comprising: S10. Obtain spectral image data of the target area from the preset database, and simultaneously obtain heart rate curve data synchronized with the spectral image acquisition period.

[0028] Specifically, the spectral image data of the target area is first retrieved from the preset database. The preset database is a database that stores the pre-acquired hyperspectral image data of the arteriovenous fistula area. The target area refers to the skin area of ​​the upper limb arteriovenous fistula that needs to be identified for failure. The spectral image data contains multiple frames of different bands, such as fistula images at 450nm, 525nm, 630nm, and 660nm. These bands correspond to the sensitive band of methemoglobin (MetHb), the reflectance characteristic band of oxyhemoglobin (HbO2), the absorption competition band between deoxyhemoglobin (Hb) and MetHb, and the strong absorption band of Hb, respectively, which can accurately reflect the spectral characteristics of different hemoglobins.

[0029] Secondly, heart rate curve data acquired synchronously with spectral image data is retrieved from a preset database. Synchronous acquisition means that the measurement time range of heart rate curve data and the acquisition time range of spectral image data completely overlap, ensuring that the two correspond one-to-one in the time dimension. Heart rate curve data records the changes in electrical signals of the heart beating during the acquisition period, including characteristic bands such as P wave, QRS complex wave, and ST segment, which can reflect changes in blood volume under different working conditions of the heart.

[0030] Next, the correspondence between each frame of the spectral image data and the time points of the heart rate curve data is established. This is achieved by binding each frame of the spectral image to a specific time point on the heart rate curve data using the timestamps from the image acquisition and heart rate measurement devices, ensuring that the corresponding spectral image can be matched based on the heart rate status later. Then, the time synchronization accuracy between the spectral image data and the heart rate curve data is verified. Typically, the time deviation is required to not exceed the minimum sampling interval of the devices. For example, when the hyperspectral camera frame rate is not less than 30, the time deviation must be less than 1 / 30 of a second. If the synchronization deviation exceeds the threshold, the system will activate a redundancy correction mechanism, including: local reconstruction of the heart rate curve based on signal interpolation, or inter-frame alignment of the spectral images using motion compensation algorithms, to reduce the risk of synchronization failure caused by device differences or slight patient movement. Furthermore, the system supports multi-device protocol adaptation and can automatically calibrate timestamps during the acquisition phase, improving data reliability in complex clinical environments. Finally, the time-synchronized spectral image data and heart rate curve data are stored and managed. They are classified and stored according to the correspondence between "image frame - heart rate time point" for easy retrieval in subsequent steps. At the same time, metadata such as the data acquisition device model and acquisition environment parameters are recorded to provide a basis for subsequent data traceability and anomaly investigation.

[0031] S11. Preprocess the spectral image data and divide the heart rate curve data into heartbeat state bands.

[0032] First, image denoising processing is performed on the acquired spectral image data of the target area. Hyperspectral images may be affected by factors such as lighting fluctuations, equipment noise, and ambient light interference during acquisition, resulting in random noise and abnormal fluctuations in image grayscale values, affecting the accuracy of subsequent feature extraction. Common denoising methods include Gaussian filtering and median filtering. By weighted averaging or sorting the grayscale values ​​within the neighborhood of image pixels, isolated noisy pixels are eliminated, while retaining effective features in the image, such as vascular texture and hemoglobin spectral characteristics. This allows the image grayscale values ​​to more accurately reflect the absorption and reflection of substances on the skin surface of the fistula area.

[0033] Furthermore, grayscale correction processing is performed on the spectral image data of the target area after noise reduction. Different detector units in a hyperspectral camera may have varying light response sensitivities, and the intensity distribution of light across different wavelengths may also be uneven. This can lead to inconsistent grayscale values ​​for the same substance at different locations or in different wavelengths of the image, making it impossible to accurately compare the spectral characteristics of hemoglobin. Grayscale correction processing establishes a grayscale correction model to map the grayscale value of each pixel in the image to a unified grayscale reference. For example, using a standard plate area with known reflectivity as a reference, the grayscale values ​​of other pixels are adjusted to ensure that the same substance has a consistent grayscale representation in the image. This ensures that the absorption and reflection intensity of light in the skin area can be accurately determined through grayscale values, thereby distinguishing different types of hemoglobin.

[0034] Spatial registration processing is performed on the spectral image data of the target area after grayscale correction. During hyperspectral image acquisition, factors such as slight equipment vibration and minor patient limb movements may cause spatial position shifts between different frames or different bands within the same frame. For example, the position of a blood vessel in one frame may not coincide with its position in the next frame, affecting the subsequent extraction and comparison of hotspot distribution features. Spatial registration processing selects feature points in the image, such as blood vessel intersections and skin edge points, and calculates the spatial transformation matrix between different images or different bands. This maps the pixels of all images or bands to the same spatial coordinate system, achieving spatial alignment of the images and ensuring that pixels at the same physical location can be accurately matched in different images or bands during subsequent analysis.

[0035] The spectral image data after image denoising, grayscale correction and spatial registration is used as the final spectral image data. This data eliminates problems such as noise interference, grayscale inconsistency and spatial offset, and can accurately reflect the spectral characteristics and spatial information of the fistula area, providing a high-quality data foundation for subsequent steps such as cardiac state band division and same state image screening.

[0036] Furthermore, the heart rate curve data is segmented into heart rate state bands. Specifically, a one-dimensional convolutional neural network (CNN) is first constructed. This network structure includes an input layer, convolutional layers, pooling layers, and fully connected layers. The input layer receives standardized heart rate curve data, that is, the fluctuation range of the heart rate curve data is standardized to the [-1,1] interval. The purpose is to eliminate the influence of differences in data units on network training and prediction, so that the network can focus more on the morphological features of the heart rate curve. The convolutional layers perform convolution operations on the input heart rate curve data through multiple convolutional kernels to extract local features of the heart rate curve, such as the rising edge of the P wave and the peak value of the QRS complex. The pooling layers downsample the feature map output by the convolutional layers, reducing the amount of data while retaining key features, thus reducing the computational complexity of the network. The fully connected layers map the feature vector output by the pooling layers to the output space, outputting the starting index of each band of the heart rate curve, such as the starting point of the P wave, the starting point of the QRS complex, and the starting point of the ST segment. These starting indices are the key basis for segmenting the heart rate state bands.

[0037] The training set for this one-dimensional CNN network consists of heart rate data with manually labeled indices for each segment of the heart rate curve. Through training with a large amount of labeled data, the network can accurately identify the starting position of different heart rate curve segments. During training, methods such as cross-validation are used to optimize network parameters and improve recognition accuracy. Then, based on the starting index output by the one-dimensional CNN network, the heart rate curve data is divided into different heart rate state segments, such as the P wave segment, corresponding to atrial systole, when the heart is supplying blood from the atria to the ventricles; the QRS complex segment, corresponding to ventricular systole, when the ventricles pump blood to the arteries; the ST segment, corresponding to pre-diastole, when the ventricles begin to relax and arterial blood flow gradually changes; and the T wave segment, corresponding to post-diastole. Each heart rate state segment corresponds to a specific working state of the heart. Different working states result in different blood flow to the heart, which in turn affects the oxygenation status of the tissue in the arteriovenous fistula area, providing a basis for subsequent screening of spectral images of the same state.

[0038] S12. Based on the divided heartbeat state bands, select spectral images in the same heartbeat state band from the spectral image data to form a set of images in the same state.

[0039] First, establish the correlation between the cardiac state bands and the acquisition time of the spectral image data. Since the correspondence between each frame of the spectral image and the time point of the heart rate curve data has been established in the data acquisition step, and the heart rate curve data has been divided into different cardiac state bands, the cardiac state band to which the acquisition time of each frame of the spectral image belongs can be determined by the time point correlation. For example, if the acquisition time of a certain frame of the spectral image corresponds to the ST segment of the heart rate curve, then that frame of the image is associated with the ST segment of the cardiac state band. Second, based on the correlation, select all spectral images whose acquisition time is within the same cardiac state band. For example, select all spectral images whose acquisition time corresponds to the QRS complex band. These images correspond to the same cardiac working state, and theoretically, the oxygenation state of the fistula area tissue is also similar, because the cardiac blood flow is stable under the same cardiac state band, the mixing ratio of arterial blood and venous blood in the fistula is relatively fixed, and the absorption and reflection characteristics of light on the skin surface are consistent.

[0040] Specifically: During the screening process, for any two spectral images in the same band but not in the same period, it is necessary to calculate their corresponding heartbeat state similarity. The calculation method is as follows: In the formula, This indicates the similarity of the heart rate states at the time points corresponding to the heart rate curves of the i-th and j-th frame spectral images, where i and j are times in the same waveband but not in the same period. and These represent the duration of the waveband in the heartbeat cycle at time i and time j, respectively. and represents the preset weighting coefficients; norm is the normalization function; , They represent the times corresponding to time i and j, respectively. and The starting duration in the time interval indicates the proportion of time in the corresponding band. For example, if the duration of the band is 10-30ms and the time interval is 12ms, then T is 20 and t is 2. and Let represent the band signals corresponding to time i and j, respectively; DTW represents the distance of the Dynamic Time Warping Algorithm (DTW). and If the signal lengths are inconsistent, linear interpolation is performed on the shorter one, followed by DTW calculation; , Both are scalars; the larger the value, the lower the similarity. Therefore, adding a negative sign indicates a negative correlation. After normalization, the larger the values ​​of the two terms, the greater the similarity of the heartbeat states. In this embodiment of the invention, a preset weighting coefficient... and The sum is 1, and the specific weighting coefficients can be set by preset parameters. and The value is 0.5, and can be adjusted by the implementer according to the actual situation. In this embodiment of the invention, the normalization function adopts the maximum and minimum value normalization function, where the maximum and minimum values ​​are the maximum and minimum values ​​in the historical data during normal operation. If the value exceeds the maximum, then the heartbeat state similarity value is set to 1. If the value is less than the minimum value, then the heartbeat state similarity value is set to 0.

[0041] Furthermore, a similarity threshold is set. If the heart rate states at two different heart rate curves are greater than the similarity threshold, it indicates that the heart rate states are consistent, and the tissue oxygenation states represented by the corresponding two frames are relatively similar. Specifically, the similarity threshold can be set to 0.9, or it can be adjusted according to actual needs. Heart rate states greater than the similarity threshold are then merged to form multiple similar sets. The corresponding spectral images are also grouped into sets in the same way to obtain spectral images within the same heart rate state band. For example, if the similarity of a, b, c, d, e, a and c, a and e, c and e, and b and d are all greater than the threshold, then the similar sets formed are ace and bd.

[0042] Furthermore, the spectral images selected from the same heartbeat state band are grouped together, with each group forming a preliminary candidate set of images in the same state. The name of the heartbeat state band to which this group belongs is recorded, such as "QRS composite wave band image group," for easy identification and retrieval in subsequent steps. Each group of spectral images in the same heartbeat state band is labeled to clearly identify the heartbeat state band to which the spectral image belongs. Labeling methods can include adding a band name suffix to the image filename or establishing an image index table to associate band information, ensuring that the heartbeat state band to which each image belongs can be quickly identified during subsequent processing, avoiding confusion between images of different bands.

[0043] Finally, the spectral images corresponding to all heartbeat state bands were screened and grouped to form multiple sets of images in the same state. The images in each set correspond to the same heartbeat state band and have similar tissue oxygenation state. This provides a unified analytical benchmark for the subsequent extraction of hotspot distribution features and calculation of differences, and reduces the interference of tissue oxygenation state differences caused by different heartbeat states on the analysis results.

[0044] S13. Extract the hotspot distribution features of each spectral image in the same state image set to obtain the hotspot distribution map corresponding to each spectral image, and determine the abnormal risk assessment value of the pixel region in each spectral image based on the difference between the hotspot distribution maps corresponding to different spectral images.

[0045] Specifically, firstly, 450nm, 525nm, 630nm, and 660nm are selected as preset feature bands, which can be adjusted according to actual needs to accurately distinguish the spectral differences of the three types of hemoglobin. For each image in the same image set, the pixel grayscale values ​​in the above four feature bands are extracted. For example, in the 450nm band, pixels with low grayscale values ​​may correspond to MetHb aggregation regions; in the 525nm band, pixels with high grayscale values ​​may correspond to normal regions with high HbO2 content. Further, an 8-neighborhood growth analysis is performed on each pixel in the spectral image, meaning the growth direction includes up, down, left, right, and four diagonals. The growth judgment condition is that the difference between the grayscale value of the pixel to be grown and the current pixel is less than a preset grayscale threshold. The growth termination condition is that the number of grown pixels reaches a preset threshold or there are no pixels that meet the grayscale difference condition. By performing growth analysis on the pixels in the spectral image, the growth region corresponding to each pixel in each spectral image is obtained.

[0046] Specifically, the preset grayscale threshold can be 20, and the preset quantity threshold can be 100, which can also be adjusted according to actual needs. Furthermore, the preset grayscale threshold and preset quantity threshold can be adaptively adjusted based on image quality and individual patient characteristics, such as skin pigmentation and vascular depth. The system can dynamically set the grayscale threshold range by analyzing the global grayscale distribution and local contrast of the image. For example, the grayscale threshold can be appropriately reduced in low-contrast areas to avoid missed features due to individual differences. Simultaneously, to improve processing efficiency, when calculating the similarity of heartbeat states, the system can use the Dynamic Time Warping (DTW) algorithm or the Piecewise Aggregate Approximation (PAA) method to reduce the dimensionality of the heart rate signal, significantly reducing computational complexity while maintaining accuracy, supporting real-time clinical processing needs.

[0047] Through growth analysis, pixels corresponding to growths with similar grayscale features and morphological patterns resembling vascular stripes are identified as hotspot pixels. Hotspot pixels represent areas with abnormal hemoglobin distribution or abnormal vascular morphology. The identification process requires calculating the distance between the farthest and nearest points within the corresponding growth region of the pixel. Specifically, the probability of hotspot feature pixels... The calculation formula is as follows: In the formula, This indicates the probability that pixel i is a hotspot feature pixel; , This represents the distance between pixel i and the farthest point in the corresponding growth region; This represents the distance between pixel i and the nearest point in the corresponding growth region. This is a normalization function. In this embodiment of the invention, the normalization function is a maximum-minimum value normalization function, where the maximum and minimum values ​​are the maximum and minimum values ​​in the historical data during normal operation. If the value exceeds the maximum value, then the probability that pixel i is a hotspot feature pixel is set to 1. If the value is less than the minimum value, then the probability that pixel i is a hotspot feature pixel is set to 0.

[0048] Specifically, the probability threshold can be set to 0.7, or adjusted as needed. If the probability of pixel i being a hotspot feature pixel is higher than the probability threshold, it is considered a hotspot pixel, and all pixels of the growth are classified as hotspot pixels and removed from the subsequent traversal range. Furthermore, all the selected hotspot pixels are combined according to their original image positions to form a hotspot distribution map that only records the positions of hotspot pixels, visually displaying the location and extent of potential risk areas in the arteriovenous fistula region.

[0049] This leads to the hotspot distribution map of each spectral image.

[0050] In some embodiments, the method further includes: classifying preset feature bands into three categories based on the spectral response characteristics corresponding to hemoglobin types: MetHb sensitive band, 450nm and HbO2 reflectance feature band, 525nm and Hb absorption feature band, and 630nm and 660nm, clearly defining the hemoglobin type corresponding to each band and avoiding spectral information interference. For each image in the image set of the same state, the pixel grayscale values ​​under the above three bands are extracted respectively. For example, the grayscale value of the 450nm band is used to determine the MetHb distribution, the grayscale value of the 525nm band is used to determine the HbO2 distribution, and the grayscale values ​​of the 630nm and 660nm bands are used to determine the Hb distribution. Within each characteristic band, hotspot pixels were selected using the neighborhood growth analysis method described above. Hotspot pixels in different bands represent different meanings: 450nm hotspot pixels correspond to potential thrombosis areas, i.e., MetHb aggregation; 525nm hotspot pixels correspond to normal blood flow areas, i.e., high HbO2 content; and 630nm and 660nm hotspot pixels correspond to blood flow obstruction areas, i.e., high Hb content. Hotspot pixels in the three bands can be mapped to the same spatial coordinate system and distinguished using different markers, such as red for 450nm hotspots, blue for 525nm hotspots, and green for Hb absorption characteristic band hotspots. The hotspot distribution map from multiple spectral images comprehensively reflects the abnormal distribution of different hemoglobins in the arteriovenous fistula area.

[0051] After obtaining hotspot distribution maps containing both hotspot and non-hotspot pixels, the degree of abnormal risk corresponding to the hotspot distribution map is calculated based on the number of hotspot pixels and the number of non-hotspot pixels; based on the degree of abnormal risk, the abnormal risk assessment value of each pixel region in each spectral image is determined.

[0052] Specifically, the abnormal risk assessment value of pixel i in the hotspot distribution map. The calculation formula is: In the formula, This represents the anomaly risk assessment value of pixel i at the same location within the same spectral band; n represents the number of categories, i.e., the number of categories. This represents the number of images contained in the k-th class set; Let i be the number of non-hotspot pixels in the k-th set. Let be the number of hotspot pixels in the k-th set, i.e., ... This is a normalization operation. It involves assessing the degree of anomaly risk. After normalization, the anomaly risk assessment value for each pixel region is calculated. The evaluation value ranges from [0,1], with values ​​closer to 1 indicating a higher risk of anomaly. At that time, set the degree of abnormal risk. The value is 1.

[0053] S14. The abnormal risk assessment values ​​that are higher than the preset abnormal risk threshold are used to form an abnormal risk distribution feature map. The abnormal risk distribution feature map is then input into a pre-trained image classification model to obtain the probability distribution of image feature types of the target region.

[0054] A preset anomaly risk threshold of 0.6 is set, which can be adjusted as needed. Values ​​higher than 0.6 are selected from the image pixel anomaly risk assessment values, and their pixel coordinates in the original image are recorded. An initial anomaly risk distribution feature map with the same size as the original spectral image is created. Pixel positions above the threshold are assigned the corresponding anomaly risk assessment value, while pixel positions below or equal to the threshold are assigned 0, visually displaying the distribution and risk level of high-risk areas.

[0055] Furthermore, morphological opening operations and other methods are used to remove regions with a value of 0 and isolated high-risk regions with an area of ​​less than 3 pixels in the initial feature map, because these regions are mostly noise. Continuous high-risk regions with a large area are retained to obtain the final abnormal risk distribution feature map. This map only contains valid abnormal risk features, providing high-quality data for model input.

[0056] The final abnormal risk distribution feature map is input into a pre-trained multi-layer convolutional neural network (CNN) classification model. The model's input layer receives the feature map, the hidden layers contain convolutional layers to extract local features such as the shape, size, and density of high-risk areas, pooling layers for downsampling to reduce data volume, and fully connected layers to integrate global features. The output layer is a probability vector of arteriovenous fistula failure type, pre-thrombotic risk, vascular stenosis risk, and vascular occlusion risk, with the sum of all probabilities equal to 1. This model is trained on a multi-center, multi-population hyperspectral arteriovenous fistula dataset, covering samples of different ages, genders, dialysis durations, and pathological types such as thrombosis, stenosis, and occlusion, ensuring good generalization ability. During training, five-fold cross-validation can be used, and the model performance can be evaluated using an independent test set. Its accuracy, sensitivity, and specificity are all superior to traditional ultrasound methods. In addition, the model supports incremental learning and can be continuously optimized based on newly acquired data to adapt to changes in the clinical environment.

[0057] The convolutional layers of the multi-layer CNN image classification model generate local risk feature maps, the pooling layers retain key features through max pooling, and the fully connected layers unfold the feature maps into one-dimensional vectors and map them to a high-dimensional space. At the same time, the model calls the historical data sample library of arteriovenous fistulas, which contains feature maps and risk probability records of the same arteriovenous fistula object at different times. The similarity between the input feature map and the features of the historical samples is calculated. If the similarity with the historical samples in the pre-thrombotic stage is high, the risk probability of the pre-thrombotic stage is tended to be increased.

[0058] Furthermore, a multi-layer CNN image classification model is used to access a multi-patient cross-sectional data sample library, containing feature maps and labeled failure types from patients of different ages, genders, and dialysis durations. The K-nearest neighbor algorithm is employed to find the K most similar samples to the input feature map, and the distribution of failure types is statistically analyzed. If the majority of samples indicate a risk of vascular stenosis, the probability of vascular stenosis risk in the input feature map is increased accordingly. Combining the comparison results of historical samples and cross-sectional samples, a comprehensive similarity is calculated according to preset weights. Specifically, historical samples have a weight of 0.6, and cross-sectional samples have a weight of 0.4, which can be adjusted as needed. Finally, the Softmax activation function is used to convert the comprehensive similarity into the risk probability of each failure type, ensuring that the probability distribution comprehensively reflects the risk of arteriovenous fistula.

[0059] Finally, the risk probability of each type of failure in the arteriovenous fistula region is output, such as the risk of prethrombosis (0.7), the risk of vascular stenosis (0.2), and the risk of vascular occlusion (0.1), forming the probability distribution of image feature types in the target region, providing doctors with quantitative references, and realizing standardized and accurate identification of arteriovenous fistula failure.

[0060] Example 2: Please see Figure 2This illustrates an embodiment of the present invention of an arteriovenous fistula failure identification system based on hyperspectral images, the system comprising: The acquisition module 20 is used to acquire spectral image data of the target area from a preset database, and simultaneously acquire heart rate curve data synchronized with the spectral image acquisition period; The segmentation module 21 is used to preprocess the spectral image data and segment the heart rate curve data into heartbeat state bands. The filtering module 22 is used to filter out spectral images in the same heartbeat state band from the spectral image data according to the divided heartbeat state bands, and form a set of images in the same state. Extraction module 23 is used to extract the hotspot distribution features of each spectral image in the same state image set, obtain the hotspot distribution map corresponding to each spectral image, and determine the abnormal risk assessment value of the pixel region in each spectral image based on the difference between the hotspot distribution maps corresponding to different spectral images. The generation module 24 is used to form an abnormal risk distribution feature map from the abnormal risk assessment values ​​that are higher than the preset abnormal risk threshold, and input the abnormal risk distribution feature map into the pre-trained image classification model to obtain the probability distribution of image feature types of the target region.

[0061] Alternatively, the transmission medium may be a wired link, such as, but not limited to, coaxial cable, fiber optic cable and digital subscriber line, or a wireless link, such as, but not limited to, wireless Fidelity (WIFI), Bluetooth and mobile device networks.

[0062] It should be noted that the device provided in the above embodiments is only an example of the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above.

[0063] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. For example, as shown... Figure 3 As shown, the computer device 30 includes: a memory 31, a processor 32, and a computer program 33 stored in the memory 31 and running on the processor 32, wherein when the processor 32 executes the computer program 33, the computer device can execute any of the aforementioned hyperspectral image-based fistula failure identification methods.

[0064] Furthermore, embodiments of the present invention also protect an apparatus that may include a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to perform the arteriovenous fistula failure identification method based on hyperspectral images provided in the embodiments of the present invention.

[0065] In this embodiment of the invention, the device can be divided into functional modules according to the above method example. For example, each module can correspond to a separate function, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and is only a logical functional division. In actual implementation, there may be other division methods.

[0066] It should be understood that the apparatus provided in this embodiment of the invention is used to perform the above-described method for identifying arteriovenous fistula failure based on hyperspectral images, and therefore can achieve the same effect as the above-described implementation method.

[0067] When using integrated units, the device may include a processing module and a storage module. When applied to a device, the processing module can be used to control and manage the device's operations. The storage module can be used to support the device in executing program code, etc. The processing module may be a processor or a controller, which can implement or execute various exemplary logic blocks, modules, and circuits as described in this disclosure. The processor may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of Digital Signal Processing (DSP) and a microprocessor, etc., and the storage module may be a memory.

[0068] In addition, the device provided in the embodiments of the present invention may specifically be a chip, component or module. The chip may include a connected processor and a memory. The memory is used to store instructions. When the processor calls and executes the instructions, the chip can execute the arteriovenous fistula failure identification method based on hyperspectral images provided in the above embodiments.

[0069] This invention also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the aforementioned method steps to implement the arteriovenous fistula failure identification method based on hyperspectral images provided in the above embodiments.

[0070] This invention also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned steps to implement the arteriovenous fistula failure identification method based on hyperspectral images provided in the above embodiments.

[0071] In this invention, the apparatus, computer-readable storage medium, computer program product, or chip provided in the embodiments are all used to execute the corresponding methods described above. Therefore, the beneficial effects they achieve can be referred to the beneficial effects in the corresponding methods described above, and will not be repeated here. Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In the embodiments provided by this invention, it should be understood that the disclosed apparatus and method can be implemented in other ways.

[0072] The device embodiments described above are merely illustrative. For example, the division of modules or units is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0073] It should also be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0074] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0075] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0076] The above content is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying arteriovenous fistula failure based on hyperspectral images, characterized in that, The method includes the following steps: Spectral image data of the target area is obtained from a preset database, and heart rate curve data synchronized with the spectral image acquisition period is obtained simultaneously. The spectral image data is preprocessed, and the heart rate curve data is divided into heartbeat state bands. Based on the divided heartbeat state bands, spectral images in the same heartbeat state band are selected from the spectral image data to form a set of images in the same state. Extract the hotspot distribution features of each spectral image in the same state image set to obtain the hotspot distribution map corresponding to each spectral image, and determine the abnormal risk assessment value of the pixel region in each spectral image based on the difference between the hotspot distribution maps corresponding to different spectral images. An abnormal risk assessment value higher than a preset abnormal risk threshold is used to form an abnormal risk distribution feature map. The abnormal risk distribution feature map is then input into a pre-trained image classification model to obtain the probability distribution of image feature types of the target region. The step of extracting hotspot distribution features from each spectral image in the same-state image set to obtain a hotspot distribution map corresponding to each spectral image includes: determining a preset feature band for extracting hotspot distribution features; analyzing the spectral images in the same-state image set based on the preset feature band; performing neighborhood growing analysis on each pixel in the spectral image, and determining hotspot pixels based on the difference between the pixel's gray value and the gray values ​​of its neighboring pixels; and combining all the determined hotspot pixels to form the hotspot distribution map corresponding to the spectral image. The step of determining the anomaly risk assessment value of pixel regions in each spectral image based on the hotspot distribution map includes: marking hotspot pixels and non-hotspot pixels according to the hotspot distribution map; calculating the anomaly risk level corresponding to the hotspot distribution map based on the number of hotspot pixels and the number of non-hotspot pixels; and determining the anomaly risk assessment value of each pixel region in each spectral image based on the anomaly risk level.

2. The method for identifying arteriovenous fistula failure based on hyperspectral images according to claim 1, characterized in that, The step of acquiring spectral image data of the target region from a preset database and simultaneously acquiring heart rate curve data synchronized with the spectral image acquisition period includes: Retrieve spectral image data of the target region from a preset database; Retrieve heart rate curve data that was synchronously acquired with the spectral image data from the preset database; Establish the correspondence between each frame of spectral image data and the time points of heart rate curve data; Verify the time synchronization accuracy between spectral image data and heart rate curve data; Store and manage the time-synchronized spectral image data and heart rate curve data.

3. The method for identifying arteriovenous fistula failure based on hyperspectral images according to claim 1, characterized in that, The preprocessing of the spectral image data includes: Image noise reduction processing is performed on the acquired spectral image data of the target area; Perform grayscale correction on the spectral image data of the target area after noise reduction; Spatial registration processing is performed on the spectral image data of the target region after grayscale correction; The spectral image data after image denoising, grayscale correction and spatial registration is used as the final spectral image data.

4. The method for identifying arteriovenous fistula failure based on hyperspectral images according to claim 1, characterized in that, The step of dividing the heart rate curve data into heartbeat state bands includes: A one-dimensional CNN network is constructed, and the heart rate curve data is processed using the one-dimensional CNN network to identify the starting index of each wave segment of the heart rate curve. The heart rate curve data is divided into heart rate state bands based on the starting index.

5. The method for identifying arteriovenous fistula failure based on hyperspectral images according to claim 1, characterized in that, Based on the defined heartbeat state bands, spectral images belonging to the same heartbeat state band are selected from the spectral image data to form a set of images with the same heartbeat state, including: Establish the correlation between the heartbeat state bands and the time of spectral image data acquisition; Based on the aforementioned correlation, all spectral images acquired at the same heartbeat state band are selected. The spectral images that are in the same heartbeat state band are grouped together. Each group of spectral images belonging to the same heartbeat state band is identified to clarify the heartbeat state band to which the spectral image belongs; Complete the screening and grouping of spectral images corresponding to all heartbeat state bands to form multiple sets of images in the same state.

6. The method for identifying arteriovenous fistula failure based on hyperspectral images according to claim 1, characterized in that, The step of forming an abnormal risk distribution feature map from abnormal risk assessment values ​​that exceed a preset abnormal risk threshold includes: From the abnormal risk assessment values ​​corresponding to the spectral image, abnormal risk assessment values ​​that are higher than the preset abnormal risk threshold are selected and mapped to the original pixel region position of the spectral image. An initial abnormal risk distribution feature map is constructed based on the abnormal risk assessment value that exceeds the preset abnormal risk threshold and its corresponding pixel region location; The initial abnormal risk distribution feature map is subjected to invalid region removal processing, and the processed abnormal risk distribution feature map is used as the final abnormal risk distribution feature map.

7. The method for identifying arteriovenous fistula failure based on hyperspectral images according to claim 1, characterized in that, The step of inputting the abnormal risk distribution feature map into a pre-trained image classification model to obtain the image feature type probability distribution of the target region includes: The abnormal risk distribution feature map is input into a pre-trained image classification model; The image classification model is used to extract and analyze the features of the abnormal risk distribution feature map, and the features of the abnormal risk distribution feature map are compared with those of historical samples in a preset historical data sample library. The image classification model calls a preset multi-patient cross-sectional data sample library and compares the input feature map with the cross-sectional sample features. By combining the comparison results of historical samples and cross-sectional samples, the pre-trained image classification model calculates the risk probability of each type of arteriovenous fistula failure. Output the risk probability of each type of failure corresponding to the arteriovenous fistula region to form the probability distribution of image feature types of the target region.

8. A fistula failure identification system based on hyperspectral images, characterized in that, The system includes the following modules: The acquisition module is used to acquire spectral image data of the target area from a preset database, and simultaneously acquire heart rate curve data synchronized with the spectral image acquisition period; The segmentation module is used to preprocess the spectral image data and segment the heart rate curve data into heartbeat state bands. The filtering module is used to filter out spectral images in the same heartbeat state band from the spectral image data according to the divided heartbeat state bands, and form a set of images in the same state. The extraction module is used to extract the hotspot distribution features of each spectral image in the same state image set, obtain the hotspot distribution map corresponding to each spectral image, and determine the abnormal risk assessment value of the pixel region in each spectral image based on the difference between the hotspot distribution maps corresponding to different spectral images. The step of extracting hotspot distribution features from each spectral image in the same-state image set to obtain a hotspot distribution map corresponding to each spectral image includes: determining a preset feature band for extracting hotspot distribution features; analyzing the spectral images in the same-state image set based on the preset feature band; performing neighborhood growing analysis on each pixel in the spectral image, and determining hotspot pixels based on the difference between the pixel's gray value and the gray values ​​of its neighboring pixels; and combining all the determined hotspot pixels to form the hotspot distribution map corresponding to the spectral image. The step of determining the anomaly risk assessment value of pixel regions in each spectral image based on the hotspot distribution map includes: marking hotspot pixels and non-hotspot pixels according to the hotspot distribution map; calculating the anomaly risk level corresponding to the hotspot distribution map based on the number of hotspot pixels and the number of non-hotspot pixels; and determining the anomaly risk assessment value of each pixel region in each spectral image based on the anomaly risk level. The generation module is used to form an abnormal risk distribution feature map from abnormal risk assessment values ​​that are higher than a preset abnormal risk threshold, and input the abnormal risk distribution feature map into a pre-trained image classification model to obtain the probability distribution of image feature types of the target region.

Images