A microthrombus feature recognition method and device based on photoacoustic imaging

Abstract

The application discloses a microthrombus feature recognition method and device based on photoacoustic imaging, and relates to the field of medical image processing.The method comprises the following steps: acquiring a photoacoustic image sequence of a to-be-detected object; splicing multiple photoacoustic images at each time point respectively; performing spatial registration on continuous blood vessel structure images at multiple time points; performing preset percentile threshold processing on the registered continuous blood vessel structure images at each time point respectively to obtain a blood vessel region mask at each time point; performing difference analysis on the blood vessel region masks at multiple time points, extracting difference features reflecting signal changes in blood vessels, and obtaining a difference feature map; performing enhanced feature recognition and weakened feature recognition on the difference feature map to obtain a microthrombus candidate region; performing morphological screening and optimization on the microthrombus candidate region to obtain a final microthrombus region, and calculating a thrombus area.The application solves the technical pain points of traditional microthrombus detection methods, such as easy missed detection, low precision and difficulty in quantitative analysis.

Description

Technical Field

[0001] This application relates to the field of medical image processing, and in particular to a method and device for identifying microthrombus features based on photoacoustic imaging. Background Technology

[0002] Microthrombus formation is a key pathological basis for various cardiovascular and cerebrovascular diseases, deep vein thrombosis, and various inflammatory responses. Unlike medium and large thrombi, microthrombi typically form in capillaries or small arteries and veins, with diameters ranging from tens to hundreds of micrometers. They are numerous, form rapidly, and are highly dynamic. Recent studies have shown that microthrombi play a crucial role in tissue microcirculatory disturbances, local ischemia, and organ dysfunction. However, due to their small size and rapid changes, the formation, detachment, and remodeling of microthrombi often occur within minutes to hours. Traditional medical imaging techniques (such as ultrasound, CT, or MRI) often face the following challenges when detecting such microscale lesions: limited spatial resolution, making it difficult to distinguish structures within tiny blood vessels; long imaging timescales, making it difficult to capture rapidly changing micro-blood flow states; and most techniques rely on exogenous contrast agents, making them unsuitable for long-term, repeated monitoring. Therefore, developing an imaging and analysis technique capable of stably detecting microthrombi under in vivo conditions and without contrast agents is of great significance for both basic medical research and clinical applications.

[0003] In recent years, photoacoustic imaging (PAI), as a novel imaging technology combining high optical contrast and deep ultrasonic penetration, has shown great potential in microvascular research. Due to the extremely high inherent light absorption properties of hemoglobin in blood, photoacoustic imaging can achieve high-resolution imaging of blood vessels without contrast agents and is widely recognized as an ideal tool for monitoring microthrombus formation. Although photoacoustic imaging has the hardware capability to capture microvascular structures, existing methods still have the following technical shortcomings that urgently need to be addressed in terms of automated microthrombus feature identification and quantitative analysis: (1) Severe interference from "spurious abnormalities" caused by physical displacement: In in vivo experiments or clinical monitoring, due to the subject's breathing, heartbeat, muscle micro-movements, or mechanical jitter of the scanning probe, there is an inevitable pixel-level spatial offset between the sequence images acquired at different time points. Existing technologies mostly use simple image subtraction methods for difference detection. This spatial misalignment will cause a large number of spurious difference signals to be generated at the edge of blood vessels, which are easily misjudged as microthrombi, resulting in a high misdiagnosis rate.

[0004] (2) Background tissue noise and signal interference from non-vascular areas: In addition to the target blood vessels, photoacoustic images also contain signals from non-vascular structures such as the skin surface and surrounding connective tissue, accompanied by laser shot noise. Existing algorithms often process the entire image area, lacking precise constraints on the analysis area, resulting in a large number of non-vascular fluctuations being incorrectly identified as thrombotic features.

[0005] (3) Systemic missed detections due to a single recognition mode: Existing recognition algorithms are usually based on a simple physical assumption that thrombosis leads to the aggregation of red blood cells, which in turn causes photoacoustic signal enhancement. However, in actual physiological processes, thrombosis may not only lead to signal enhancement (accumulation type), but may also cause local blood loss due to complete blockage of blood vessels, which in turn causes signal disappearance or significant weakening (obstruction type). If existing schemes only detect "enhanced signals", it will cause serious systemic missed detections.

[0006] (4) High environmental sensitivity and lack of adaptive quantization capability: The intensity of photoacoustic signals is easily affected by the fluctuation of laser pulse energy, acoustic coupling state and imaging depth. Existing technologies mostly use fixed grayscale thresholds for segmentation. This "one-size-fits-all" method will have a significant decrease in accuracy when faced with fluctuations in hardware environment or individual differences, and it is difficult to achieve automated quantitative comparison of cross-batch data. Summary of the Invention

[0007] The purpose of this application is to provide a method and device for microthrombus feature recognition based on photoacoustic imaging, which can improve the accuracy of microthrombus feature recognition.

[0008] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for identifying microthrombus features based on photoacoustic imaging, including: Acquire a photoacoustic image sequence of the object under test; the photoacoustic image sequence includes multiple consecutive frames of photoacoustic images at multiple time points; The multiple frames of photoacoustic images at each time point are stitched together to obtain continuous vascular structure images at each time point; Spatial registration is performed on continuous vascular structure images at multiple time points to obtain registered continuous vascular structure images at each time point; The registered continuous vascular structure images at each time point are processed by a preset percentile threshold to obtain the vascular region mask at each time point. Differential analysis was performed on vascular region masks at multiple time points to extract differential features reflecting changes in intravascular signals and obtain differential feature maps. Enhanced and weakened feature recognition are performed on the differential feature map to obtain microthrombus candidate regions; The candidate microthrombus regions are morphologically screened and optimized to obtain the final microthrombus regions, and the thrombus area is calculated.

[0009] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described microthrombus feature recognition method based on photoacoustic imaging.

[0010] According to the specific embodiments provided in this application, this application has the following technical effects: By stitching together photoacoustic image sequences from multiple time points, continuous vascular structures can be completely restored, overcoming the limitations of single-frame image field of view; spatial registration effectively eliminates spatial deviations in vascular images at different time points, ensuring spatial consistency in subsequent analysis; preset percentile threshold processing can accurately segment vascular regions and reduce interference from non-vascular regions; differential analysis can keenly capture the signal change characteristics within blood vessels, providing a core basis for microthrombus identification; enhanced and weakened dual-mode feature recognition can comprehensively cover different signal manifestations of microthrombi, reducing the false negative rate; morphological screening and optimization further eliminate artifact interference, improving the accuracy and completeness of microthrombus region identification; combined with the thrombus area calculation function, it can provide quantitative assessment indicators for clinical use. The overall solution requires no invasive operation, has high identification efficiency and excellent accuracy, and solves the technical pain points of traditional microthrombus detection methods, such as easy false negatives, low accuracy, and difficulty in quantitative analysis, providing reliable technical support for the early diagnosis and disease monitoring of thrombosis-related diseases. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is an application environment diagram of a microthrombus feature recognition method based on photoacoustic imaging in one embodiment of this application.

[0013] Figure 2 This is a flowchart illustrating a microthrombus feature recognition method based on photoacoustic imaging, provided as an embodiment of this application.

[0014] Figure 3 A framework diagram for identifying microthrombus features is provided for one embodiment of this application.

[0015] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] To address the shortcomings of existing technologies, such as low accuracy, poor robustness, and high risk of missed detection, this application proposes a comprehensive R&D concept of "spatial registration + mask constraint + bidirectional statistical recognition." Without requiring the injection of exogenous contrast agents, hemoglobin is used as an endogenous contrast agent, and computer vision algorithms are employed to mine deep pathological features in time-series images. Firstly, sub-pixel-level registration solves the "motion" problem; secondly, skeleton-guided mask logic addresses the "background noise" problem; the most crucial breakthrough lies in the introduction of bidirectional difference recognition logic based on statistical distribution, simultaneously incorporating both "signal enhancement" and "signal weakening" pathological manifestations into the monitoring scope. The resulting automated recognition method aims to provide a high-precision, high-stability, and quantifiable technical tool for biomedical research and clinical microcirculation assessment, thereby assisting doctors and researchers in accurately determining the location, evolution, and drug intervention effects of microthrombi.

[0018] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] The microthrombus feature recognition method based on photoacoustic imaging provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be set up independently, integrated into server 102, or placed in the cloud or on another server. Terminal 101 can send a photoacoustic image sequence of the object under test to server 102. After receiving the photoacoustic image sequence, server 102 performs microthrombus feature recognition to obtain the final microthrombus region and calculates the thrombus area. Server 102 can then return the final microthrombus region and thrombus area to terminal 101.

[0020] The terminal 101 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 102 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0021] In one exemplary embodiment, such as Figure 2 As shown, a microthrombus feature recognition method based on photoacoustic imaging is provided. This method is executed by a computer device (such as a computer, server, or dedicated image processing hardware) and aims to solve the problem of difficult microthrombus recognition caused by biological tissue background interference and signal fluctuations in photoacoustic imaging. In the embodiments of this application, this method is applied to... Figure 1 Taking server 102 as an example, the explanation includes the following steps 201 to 208.

[0022] Step 201: Obtain the photoacoustic image sequence of the object under test. The photoacoustic image sequence includes multiple consecutive frames of photoacoustic images at multiple time points. Each frame of the photoacoustic image sequence is sequentially processed by grayscale normalization, Gaussian filtering, intensity normalization, and invalid region cropping.

[0023] Specifically, computer equipment acquires photoacoustic signals from the same site at different times or different pathological stages using a photoacoustic imaging system, and reconstructs them into photoacoustic images in the form of a pixel matrix. The photoacoustic image sequence includes a reference image (usually in a normal blood flow state or an initial state) and at least one frame of the image to be analyzed.

[0024] It should be noted that, due to the limited scanning field of view in practical applications, photoacoustic imaging systems often only cover a local vascular area in a single scan. Therefore, under most experimental conditions, the photoacoustic image is not a single complete image, but rather consists of multiple scan images with spatial overlap.

[0025] Furthermore, photoacoustic images possess the following engineering characteristics: photoacoustic signal intensity is significantly affected by minute fluctuations in laser energy; changes in probe angle and acoustic coupling state introduce non-pathological grayscale fluctuations; and invalid signals often exist in image edge regions and coupling medium regions. Based on these characteristics, preprocessing is required before proceeding to subsequent analysis. Preprocessing operations include, but are not limited to: removing NaN values ​​from the photoacoustic image; grayscale normalization to reduce overall intensity differences between different scanning batches; using Gaussian filtering and other smoothing methods to suppress high-frequency noise; intensity normalization of the photoacoustic signal to reduce the impact of laser energy fluctuations; and cropping invalid regions such as probe edge regions and coupling medium regions from the image. These preprocessing steps ensure that the photoacoustic image used in subsequent analysis has a more stable signal-to-noise ratio and intensity distribution, providing reliable input for subsequent stitching, registration, and difference analysis.

[0026] Step 202: The multiple frames of photoacoustic images at each time point are stitched together to obtain a continuous vascular structure image at each time point.

[0027] In a specific application example, to address the problem of limited field of view in a single scan, this application automatically stitches together multiple photoacoustic images after preprocessing to obtain a large-field photoacoustic image covering continuous vascular structures. In this step, this application does not employ manual point selection or fixed offset stitching methods, but rather automatically estimates the spatial offset relationship by analyzing the overlapping areas between adjacent images. Step 202 includes steps 21 to 25.

[0028] Step 21: For the i-th stitching at any given time point, extract the overlapping region between the current stitched image and the current image to be stitched at that time point, and calculate the similarity distribution between the overlapping regions. If i > 0, the current stitched image at the first stitching is the first frame photoacoustic image at that time point, and the current image to be stitched is the next frame photoacoustic image of the current stitched image.

[0029] Step 22: Determine the pixel-level offsets in the horizontal and vertical directions between the current image to be stitched and the current image to be stitched based on the similarity distribution. Specifically, determine the pixel-level offsets in the horizontal and vertical directions between the two images based on the position of the maximum similarity value.

[0030] Step 23: Perform a geometric translation on the current image to be stitched according to the pixel-level offset, so that the current image to be stitched is spatially aligned with the current image to be stitched.

[0031] Step 24: Calculate the pixel weighting coefficients of the aligned overlapping areas, and stitch the current stitched image with the aligned current image to be stitched according to the pixel weighting coefficients. At the same time, use a linear transition algorithm to eliminate stitching gaps to obtain the current stitched image at the (i+1)th stitching.

[0032] Step 25: If the current image to be stitched is the last photoacoustic image at the time point, then the current stitched image at the (i+1)th stitching is taken as the continuous vascular structure image at the time point, that is, a continuous large field-of-view image containing the complete vascular pathway is generated to ensure that the microthrombus analysis is performed under the complete anatomical structure; otherwise, the (i+1)th stitching is performed.

[0033] Through the above-described stitching steps, this application can obtain complete and continuous vascular photoacoustic imaging results without introducing obvious stitching artifacts, providing a large-field structural basis for subsequent microthrombus analysis.

[0034] Step 203: Spatial registration is performed on continuous vascular structure images at multiple time points to obtain registered continuous vascular structure images at each time point.

[0035] After stitching together multiple photoacoustic images to obtain continuous vascular structure images, this application further addresses the spatial inconsistency between photoacoustic images acquired at different time points or under different experimental conditions by performing a high-precision spatial registration operation.

[0036] In photoacoustic microthrombus detection, image registration is not for macroscopic structural alignment, but rather a crucial prerequisite that directly determines whether subsequent pixel-level difference analysis has true pathological significance. This is because the spatial scale corresponding to microthrombi in photoacoustic images is typically only tens to hundreds of micrometers, occupying a very small number of pixels in the image. In this case, even a spatial misalignment of only 1 to 2 pixels can be magnified into a high-intensity abnormal area in the difference map, resulting in a large number of false thrombus detection results. However, in actual photoacoustic imaging, images acquired at different time points almost inevitably contain the following sources of spatial deviation: slight initial position shifts during mechanical scanning; angular differences caused by changes in probe installation or fixation; slight tissue displacement during imaging; and accumulated mechanical errors during repeated scanning.

[0037] The aforementioned deviations typically manifest as pixel-level translations, accompanied by slight rotations or affine changes, but their impact on microthrombus detection is decisive, requiring computer equipment to eliminate geometric misalignments between images. Based on these engineering realities, this application does not employ conventional registration methods based on feature points (such as corner points, SIFT, SURF) or mutual information. The reasons include: firstly, the overall texture of photoacoustic vascular images is relatively sparse, with a limited number of stable feature points, making reliable feature matching difficult; secondly, there are often laser energy fluctuations and acoustic coupling differences between photoacoustic images at different time points, resulting in grayscale distributions that do not meet the brightness consistency assumption, thus leading to insufficient stability of methods based on direct grayscale matching.

[0038] Therefore, this application adopts an image registration strategy based on Normalized Cross-Correlation (NCC). This method normalizes the mean and variance of the template region and the search region, making the correlation calculation invariant to changes in overall brightness, and effectively suppressing the intensity drift problem commonly found in photoacoustic imaging. Step 203 includes steps 31 to 35.

[0039] Step 31: Use the continuous vascular structure image at the first time point as the reference image, and extract salient features (such as vascular bifurcation points) from the reference image to obtain a matching template. t The matching template typically occupies about 10% to 20% of the original image size, which is sufficient to contain vascular texture information while avoiding the introduction of large areas of low signal-to-noise ratio background regions into the correlation calculation. Significant features can be extracted through manual annotation.

[0040] Step 32: The correlation coefficient distribution between the matching templates in the images to be registered is calculated using a normalized cross-correlation algorithm to obtain a correlation coefficient matrix. The images to be registered are continuous vascular structure images at all time points except the first time point.

[0041] Specifically, using the formula Calculate the cross-correlation coefficient. Among them, For cross-correlation coefficients, For the image to be registered ( x , y The pixel value at position ) To match the template Pixel value at the location, For the image to be registered ( u , v The mean at position ) For matching template The mean at the location, W It is a set of pixels.

[0042] Step 33: Determine the pixel-level coordinate offsets of the image to be registered relative to the reference image in the X and Y directions based on the correlation coefficient matrix. Specifically, the pixel-level coordinate offsets of the image to be registered relative to the reference image in the X and Y directions are obtained by searching for the position of the maximum peak in the correlation coefficient matrix.

[0043] Step 34: Based on the pixel-level coordinate offset, transform the image to be registered so that it is mapped to the coordinate system of the reference image, obtaining a preliminary registered image at each time point. Specifically, based on the pixel-level coordinate offset corresponding to the maximum peak position, construct a corresponding geometric transformation model, and perform translation or affine transformation on the image to be registered so that it is mapped to the coordinate system of the reference image.

[0044] Step 35: Perform effective overlapping region cropping on the preliminary registered images at each time point, retaining only the spatially corresponding pixel regions, to obtain continuous vascular structure images after registration at each time point.

[0045] It is important to note that after registration, this application does not directly use the transformed entire image for subsequent difference analysis. Instead, it further crops the effective overlapping areas of the two images, retaining only the spatially corresponding pixel regions. This operation avoids invalid pixels caused by boundary padding, interpolation resampling, etc., from being misjudged as abnormal signals.

[0046] Through the above registration process, this application can achieve sub-pixel-level photoacoustic image alignment even in the presence of laser energy fluctuations, acoustic coupling differences, and slight scanning displacement, providing a reliable spatial consistency guarantee for subsequent microthrombus detection based on pixel differences.

[0047] Furthermore, without altering the core objective of "spatially aligning photoacoustic images from different times to ensure the effectiveness of pixel-level difference analysis," other image registration methods can be used as alternatives. These include: image registration methods based on mutual trust; other template matching methods based on correlation coefficients or similarity metrics; image registration methods based on affine transformations or non-rigid transformation models; and, where training conditions are available, image registration models based on machine learning or deep learning. All of these alternatives can achieve spatial alignment of photoacoustic images from different times; their differences lie only in the specific similarity calculation methods or transformation models, and all can serve as equivalent substitutes for the spatial registration steps in this application.

[0048] Step 204: Perform preset percentile threshold processing on the registered continuous vascular structure images at each time point to obtain the vascular region mask at each time point.

[0049] To eliminate signal interference from extravascular tissues (such as the skin surface and surrounding connective tissue), this application further constructs a vascular region mask to strictly limit the analysis scope for microthrombus feature recognition. In the microthrombus detection task, the vascular region mask is not used to simply display vascular structures, but rather as a strong constraint for subsequent pixel-level difference analysis and abnormal region screening. If the vascular region extraction is inaccurate, background noise, tissue signal fluctuations, and boundary artifacts in non-vascular regions will inevitably be introduced into the microthrombus detection process, thereby significantly increasing the false detection rate.

[0050] In existing technologies, blood vessel extraction typically employs grayscale segmentation methods based on a single threshold. However, in photoacoustic imaging scenarios, this type of method has significant shortcomings: on the one hand, the photoacoustic signal intensity varies greatly between different blood vessel branches, making it difficult for a single threshold to simultaneously cover both the main blood vessels and low-signal micro-branches; on the other hand, photoacoustic images often contain background structures with similar grayscale values ​​to blood vessels, making it easy for simple threshold segmentation to misidentify noisy regions as blood vessels.

[0051] In a specific application example, the preset percentile threshold is determined in advance using a machine learning algorithm. That is, the machine learning model is trained in advance using continuous vascular structure image samples and different percentile thresholds, and the optimal percentile threshold is selected as the optimal percentile threshold.

[0052] In another specific application example, a skeleton-guided two-stage vascular region mask construction strategy can also be adopted.

[0053] The first stage involves high percentile thresholding of the registered continuous vascular structure image, retaining only pixels in the high-intensity range of the grayscale distribution. This aims to reduce background noise interference in the skeleton extraction process. Subsequently, morphological denoising and thinning are performed on the resulting binary image to extract the central skeleton structure of the vascular candidate region. To avoid short skeletons formed by random noise or local highlight artifacts being misidentified as vascular structures, a further screening mechanism based on the skeleton's principal axis length is introduced, retaining only skeleton branches whose length exceeds a preset pixel threshold. This ensures that the retained skeleton structure originates from real blood vessels rather than isolated noise. It is important to emphasize that the skeleton structure extracted in this stage is not directly used as the final vascular mask. Directly using the binary result of this stage as the vascular region will inevitably result in the loss of vascular boundary regions and some low-signal branches, thus limiting the coverage of subsequent microthrombus detection.

[0054] The second stage involves binarizing the registered continuous vascular structure images using different percentile thresholds to obtain a set of candidate regions containing vascular boundary information. Subsequently, only connected regions that spatially intersect with the skeleton structures extracted in the first stage are retained as the final vascular region mask.

[0055] By employing the aforementioned skeleton-guided strategy, this application preserves the integrity of the real vascular structure to the greatest extent possible while eliminating background noise, providing a reliable spatial constraint for subsequent differential analysis.

[0056] Furthermore, without altering the fundamental technical principle of "limiting the analysis scope to the real vascular structure to exclude interference from non-vascular regions," other methods for vascular region extraction or limitation can also be employed. These include: methods based on global or local threshold segmentation combined with morphological processing; vascular extraction methods based on region growing or connectivity analysis; vascular region recognition methods based on traditional machine learning feature classification; and, when sample conditions are available, deep learning-based vascular segmentation networks. All of these methods can be used to obtain vascular region masks; although their implementations differ, their functional purpose is the same, and they all represent alternative implementations to the vascular region limitation step of this application.

[0057] Step 205: Perform differential analysis on the vascular region masks at multiple time points, extract differential features reflecting changes in intravascular signals, and obtain a differential feature map. The differential feature map includes the relative change of each pixel, representing the change curve between consecutive time points.

[0058] After constructing a vascular region mask, this application performs differential analysis on photoacoustic images at different time points or under different conditions within the range defined by the vascular region mask to extract differential features reflecting changes in intravascular signals. It should be noted that under photoacoustic imaging conditions, there are often significant differences in the basic photoacoustic signal intensity between different vascular branches. This difference may originate from differences in vessel diameter, or from differences in local blood oxygenation status, blood flow velocity, and tissue optical properties. Therefore, if anomaly detection is directly based on absolute grayscale differences, it often leads to the amplification of anomalies in high-signal main vascular regions, while anomalies in low-signal small branch regions are masked. Based on the above problems, this application adopts a formula... Calculate the mask in the blood vessel region of the two images after registration. The relative change within the range. Among them, for( x , y The relative change at the location. and Within the mask area of ​​the blood vessel region in two frames of images ( x , y The pixel value at the given location.

[0059] Specifically, by normalizing the registered photoacoustic images, the relative signal changes between images at different time points are calculated, ensuring that the difference features primarily reflect changes caused by time or state variations, rather than intensity differences inherent in spatial location. Subsequently, the relative difference map is combined with a vascular region mask, retaining only the difference features within the blood vessels, thus obtaining a difference feature map for microthrombus identification.

[0060] Furthermore, without altering the core objective of "analyzing changes in intravascular photoacoustic signals based on time or state changes," other differential analysis methods can also be employed, such as: signal change analysis methods based on absolute or normalized differences; change detection methods based on time series modeling; anomaly detection methods based on statistical or probabilistic models; and change region identification methods based on cluster analysis or classification models. All of these approaches can identify regions of signal change within blood vessels; their differences lie in the feature construction methods or judgment rules, and all can serve as alternatives to the differential analysis steps in this application.

[0061] Step 206: Perform enhanced feature recognition and weakened feature recognition on the differential feature map to obtain microthrombus candidate regions.

[0062] Under photoacoustic imaging conditions, the formation of microthrombi does not always manifest as a unidirectional signal change. Experiments show that microthrombi can lead to local blood stasis and increased hemoglobin concentration, thus appearing as signal enhancement in photoacoustic images; or they can cause blood flow obstruction and decreased local perfusion, resulting in a significant decrease in photoacoustic signal. If only a unidirectional thresholding strategy is used, detecting only signal-enhanced areas will inevitably miss microthrombi that appear as signal-enhanced areas; and vice versa. After obtaining the differential feature map within the vascular region, this application does not use a single fixed threshold, but rather performs dynamic identification based on the statistical distribution (such as percentiles) of the differential feature map.

[0063] In a specific application example, the microthrombus candidate region includes a signal-enhancing region and a signal-degrading region. Step 206 includes steps 61 and 62.

[0064] Step 61: Identify that the relative change in the difference feature map is greater than a set enhanced change threshold (e.g., ...). The signal enhancement region is obtained by analyzing the pixels of the quantile. Its technical significance lies in capturing the abnormal enhancement of photoacoustic signals caused by blood stasis and red blood cell aggregation.

[0065] Step 62, identify that the relative change in the difference feature map is less than a set weakened change threshold (e.g., The pixel with quantiles is used to obtain the signal attenuation region. Its technical significance lies in capturing the signal loss caused by complete blockage leading to no blood perfusion.

[0066] The threshold values ​​for enhancement and reduction are not fixed but adaptively vary with the differential distribution, thus adapting to data changes in different experimental batches, different vascular structures, and different imaging conditions.

[0067] Step 207: Morphological screening and optimization are performed on the candidate microthrombus regions to obtain the final microthrombus regions, and the thrombus area is calculated.

[0068] The microthrombus candidate region may still contain spurious abnormal regions caused by factors such as noise, minor vascular deformation, and registration errors. Based on the above engineering realities, this application further introduces a multi-stage morphological screening and optimization mechanism to impose layer-by-layer constraints on the microthrombus candidate region.

[0069] In a specific application example, step 207 includes steps 71 to 74.

[0070] Step 71: Perform connected component analysis on the microthrombus candidate regions and remove connected components smaller than the first preset area threshold to obtain preliminary microthrombus regions, thereby quickly reducing the number of candidate regions.

[0071] Step 72: Perform morphological closing operations on the preliminary microthrombus regions to repair regional fragmentation issues and obtain continuous microthrombus regions. Specifically, perform structural optimization operations such as morphological closing operations on the screened abnormal regions to repair regional fragmentation issues caused by difference threshold segmentation or slight registration errors, so that the real microthrombus regions exhibit a relatively continuous structure in space.

[0072] Step 73: Perform connected component analysis on the continuous microthrombus region and remove connected components smaller than the second preset area threshold to obtain the final microthrombus region. That is, only the region that meets the microthrombus imaging characteristics in both spatial scale and continuity is retained as the final identification result. The first preset area threshold is greater than the second preset area threshold.

[0073] Furthermore, without altering the objective of "removing noisy regions and retaining physiologically significant anomalous regions," other region optimization and screening strategies can be employed. These include: using different combinations of morphological operations to repair and optimize regions; screening based on region shape, compactness, or topological features; and using rule-based or learning-based region discrimination methods. All of these solutions functionally achieve the screening and optimization of anomalous regions and represent alternative implementations to the post-processing steps of this application.

[0074] Step 74: Calculate the thrombus area based on the total number of pixels and imaging resolution within the final microthrombus region.

[0075] Step 208: In the photoacoustic image sequence, signal-enhancing regions are marked with a first preset color (e.g., red outlines), and signal-attenuating regions are marked with a second preset color (e.g., blue outlines), and then displayed. Specifically, the signal-enhancing and signal-attenuating regions marked with the colors are superimposed on the original photoacoustic image on a computer monitor.

[0076] In addition, if it is a continuous scanning sequence, a "time-thrombus area" trend curve is generated to assess the growth or ablation rate of the thrombus.

[0077] In another exemplary embodiment, such as Figure 3 As shown, the photoacoustic image sequence was divided into a control group and an experimental group. The control group was subjected to binarization, skeleton refinement, and filtering based on the skeleton's principal axis length to obtain the vascular mask data. Both the control and experimental groups underwent preprocessing, registration, difference analysis, feature recognition, and thrombus labeling.

[0078] In summary, this application utilizes computer equipment to process photoacoustic imaging data, thereby achieving a method for automatic identification and quantitative analysis of microthrombi in living organisms. This method uses photoacoustic images of the same subject acquired at different time points or under different experimental conditions as input data, and achieves stable identification of microthrombi through the following technical path: 1. Precise spatial alignment of multi-frame images: For photoacoustic image sequences at different time points or under different scanning states, sub-pixel-level registration is performed using algorithms such as normalized cross-correlation. By calculating the similarity of overlapping areas and compensating for spatial displacement, a consistent geometric benchmark is provided for subsequent difference analysis.

[0079] 2. Adaptive mask construction for vascular regions: A strategy combining skeleton extraction and connectivity analysis is adopted to automatically identify the main trunk and branch structures of blood vessels from photoacoustic images and generate vascular region masks to ensure that subsequent thrombus identification and analysis are strictly limited to the inside of the blood vessel lumen and to eliminate background noise interference.

[0080] 3. Thrombosis region identification based on bidirectional difference features: Calculate the relative difference distribution between the image to be analyzed and the baseline image, and simultaneously identify two types of abnormal features based on adaptive statistical thresholds (such as percentiles): Signal enhancement region: corresponding to signal enhancement (highlighted area) caused by blood stasis and red blood cell aggregation; Signal attenuation region: corresponding to signal loss (blockage area) caused by complete obstruction of blood flow and no perfusion.

[0081] 4. Morphological optimization and quantitative output: Multi-stage morphological processing (such as area filtering, closing operation filling, etc.) is performed on the initially identified abnormal areas to eliminate random noise and optimize the region boundaries. Finally, quantitative indicators such as the area and location of microthrombi are calculated based on the identification results and presented in color-coded visualization.

[0082] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores photoacoustic image sequences of the object under test. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a microthrombus feature recognition method based on photoacoustic imaging.

[0083] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0084] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0085] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0086] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0087] In this application, all actions to acquire signals, information, or data are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with the authorization granted by the owner of the relevant device.

[0088] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0089] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, etc., and are not limited to these.

[0090] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0091] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for identifying microthrombi based on photoacoustic imaging, characterized in that, The microthrombus feature identification method based on photoacoustic imaging includes: Acquire a photoacoustic image sequence of the object under test; the photoacoustic image sequence includes multiple consecutive frames of photoacoustic images at multiple time points; The multiple frames of photoacoustic images at each time point are stitched together to obtain continuous vascular structure images at each time point; Spatial registration is performed on continuous vascular structure images at multiple time points to obtain registered continuous vascular structure images at each time point; The registered continuous vascular structure images at each time point are processed by a preset percentile threshold to obtain the vascular region mask at each time point. Differential analysis was performed on vascular region masks at multiple time points to extract differential features reflecting changes in intravascular signals and obtain differential feature maps. Enhanced and weakened feature recognition are performed on the differential feature map to obtain microthrombus candidate regions; The candidate microthrombus regions are morphologically screened and optimized to obtain the final microthrombus regions, and the thrombus area is calculated.

2. The microthrombus feature identification method based on photoacoustic imaging according to claim 1, characterized in that, Before stitching together the multi-frame photoacoustic images at each time point, the microthrombus feature recognition method based on photoacoustic imaging further includes: Each frame of the photoacoustic image sequence is sequentially subjected to grayscale normalization, Gaussian filtering, intensity normalization, and invalid region cropping.

3. The microthrombus feature identification method based on photoacoustic imaging according to claim 1, characterized in that, Multiple frames of photoacoustic images at each time point are stitched together to obtain continuous vascular structure images at each time point, including: For the i-th stitching at any given time point, the overlapping region between the current stitched image and the current image to be stitched at that time point is extracted, and the similarity distribution between the overlapping regions is calculated; i>0, the current stitched image at the time point of the first stitching is the first frame photoacoustic image at that time point, and the current image to be stitched is the next frame photoacoustic image of the current stitched image. The pixel-level offsets between the current stitched image and the current image to be stitched are determined in the horizontal and vertical directions based on the similarity distribution. The current image to be stitched is geometrically translated according to the pixel-level offset, so that the current image to be stitched is spatially aligned with the current image to be stitched. Calculate the pixel weighting coefficient of the aligned overlapping area, and stitch the current stitched image with the aligned current image to be stitched according to the pixel weighting coefficient. At the same time, use a linear transition algorithm to eliminate stitching gaps to obtain the current stitched image at the (i+1)th stitching. If the current image to be stitched is the last photoacoustic image at the time point, then the current stitched image at the (i+1)th stitching is taken as the continuous vascular structure image at the time point; otherwise, the (i+1)th stitching is performed.

4. The microthrombus feature identification method based on photoacoustic imaging according to claim 1, characterized in that, Spatial registration is performed on continuous vascular structure images at multiple time points to obtain registered continuous vascular structure images at each time point, including: The continuous vascular structure image at the first time point is used as the reference image, and significant features are extracted from the reference image to obtain the matching template; The correlation coefficient distribution between the matching templates in the images to be registered is calculated using a normalized cross-correlation algorithm to obtain a correlation coefficient matrix; the images to be registered are continuous vascular structure images at all time points except the first time point. The pixel-level coordinate offsets of the image to be registered relative to the reference image in the X and Y directions are determined based on the correlation coefficient matrix. Based on the pixel-level coordinate offset, the image to be registered is transformed so that it is mapped to the coordinate system of the reference image, thus obtaining the preliminary registration image at each time point. The effective overlapping regions of the preliminary registered images at each time point are cropped, and only the spatially corresponding pixel regions are retained to obtain continuous vascular structure images after registration at each time point.

5. The microthrombus feature identification method based on photoacoustic imaging according to claim 1, characterized in that, The preset percentile threshold is determined in advance using a machine learning algorithm.

6. The microthrombus feature recognition method based on photoacoustic imaging according to claim 1, characterized in that, The difference feature map includes the relative change of each pixel.

7. The microthrombus feature recognition method based on photoacoustic imaging according to claim 6, characterized in that, The microthrombus candidate region includes signal-enhancing regions and signal-attenuating regions; Enhanced and weakened feature recognition are performed on the differential feature map to obtain microthrombus candidate regions, including: Identify pixels in the difference feature map whose relative change is greater than a set enhancement change threshold to obtain the signal enhancement region; Identify pixels in the difference feature map whose relative change is less than a set attenuation threshold to obtain the signal attenuation region.

8. The microthrombus feature recognition method based on photoacoustic imaging according to claim 7, characterized in that, The microthrombus feature recognition method based on photoacoustic imaging also includes: In the photoacoustic image sequence, a first preset color is used to mark the signal enhancement area, and a second preset color is used to mark the signal reduction area, and these areas are displayed.

9. The microthrombus feature identification method based on photoacoustic imaging according to claim 1, characterized in that, The candidate microthrombus regions are morphologically screened and optimized to obtain the final microthrombus regions, and the thrombus area is calculated, including: Connectivity analysis is performed on the candidate microthrombus regions, and connected regions smaller than a first preset area threshold are removed to obtain preliminary microthrombus regions; Morphological closure processing is performed on the initial microthrombus region to repair the region breakage problem and obtain a continuous microthrombus region; Connectivity analysis is performed on the continuous microthrombus region, and connected components smaller than the second preset area threshold are removed to obtain the final microthrombus region; the first preset area threshold is greater than the second preset area threshold. The thrombus area is calculated based on the total number of pixels and imaging resolution within the final microthrombus region.

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the microthrombus feature recognition method based on photoacoustic imaging as described in any one of claims 1-9.

Images