A registration method for multimodal retinal blood flow image processing

Abstract

The present invention discloses a registration method for multimodal retinal blood flow image processing, which relates to the field of image processing technology and includes the following steps: first, receiving retinal blood flow image data from different modalities to prepare for subsequent data processing and registration, and after receiving retinal blood flow images of different modalities, performing image preprocessing to reduce the differences between different modalities and ensure that the images are analyzed and registered under the same standard. The present invention predicts the dynamic change trend of blood flow through a support vector machine model, dynamically adjusts the similarity measurement threshold, flexibly responds to local natural deformation, and avoids registration failure caused by blood flow fluctuations, thereby improving the robustness and accuracy of multimodal image registration, especially in areas with significant blood flow changes, achieving more accurate spatial alignment, and providing high-quality image data support for retinal disease diagnosis and pathological analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a registration method for multi-modal retinal blood flow image processing. Background Art

[0002] Multimodal retinal blood flow image processing registration refers to the process of aligning and registering retinal blood flow images from different imaging technologies (such as OCT, fluorescent angiography, retinal blood flow imaging, etc.) in order to fuse images from different modalities into a unified coordinate system. This process is of great significance for obtaining more comprehensive retinal blood flow information and improving diagnostic accuracy. Since images produced by different imaging methods may have differences in size, viewing angle, resolution, etc., registration technology ensures the correspondence between images through mathematical and computational methods (such as mutual information, gradient, feature matching, etc.), thereby realizing the integration of different imaging data, providing a clearer and more comprehensive retinal blood flow dynamic information, thereby helping doctors to make more accurate diagnoses and treatments.

[0003] The prior art has the following deficiencies:

[0004] In multimodal retinal blood flow image registration, existing technologies usually evaluate the degree of image matching by setting a fixed similarity threshold. For example, when the mutual information value exceeds a certain set threshold, the two images are considered to have been successfully aligned. However, due to the dynamic characteristics of blood flow (such as vascular expansion, contraction, and changes in blood flow velocity), traditional similarity measurement methods (such as mutual information, mean square error, etc.) are often difficult to fully capture the impact of these dynamic changes on image features. Therefore, when faced with situations where the dynamic changes of blood flow are significant, the fixed similarity threshold standard may not accurately reflect the true matching relationship between images, resulting in a decrease in registration accuracy. In particular, when the blood flow changes dramatically or the local vascular structure changes significantly, the constant similarity threshold may ignore the natural deformation caused by the blood flow change, ultimately leading to registration failure or mismatch.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a registration method for multimodal retinal blood flow image processing, which accurately predicts the dynamic change trend of blood flow in different regions through a support vector machine model, and dynamically adjusts the similarity measurement threshold for areas with drastic blood flow changes, so as to flexibly respond to local natural deformations and avoid registration failures caused by dynamic changes such as blood flow fluctuations, vascular expansion or contraction. By introducing a dynamically adjusted threshold standard, the registration algorithm has stronger robustness and adaptability, and more accurate spatial alignment and structural matching are achieved in multimodal images, especially in local areas with significant blood flow changes, which significantly improves the overall accuracy and reliability of registration, and provides more accurate image data support for subsequent retinal disease diagnosis and pathological analysis, so as to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a registration method for multimodal retinal blood flow image processing, comprising the following steps:

[0008] First, retinal blood flow image data from different modalities are received to prepare for subsequent data processing and registration. After receiving retinal blood flow images of different modalities, image preprocessing is performed to reduce the differences between different modalities and ensure that the images are analyzed and registered under the same standard.

[0009] Extract blood flow information from each pre-processed retinal blood flow image to provide a basis for subsequent blood flow dynamic characteristics analysis. Filter out key features of blood flow dynamic changes from the extracted blood flow information, analyze the filtered features in detail, and quantify the degree of change in blood flow dynamic characteristics.

[0010] The analyzed and quantified blood flow dynamic characteristics are input into the pre-trained support vector machine model to predict the blood flow dynamic change trend in different areas;

[0011] The similarity measurement threshold is dynamically adjusted according to the dynamic characteristics of blood flow predicted by the support vector machine model. For areas with drastic blood flow changes, the deformation degree of the area is judged by the prediction results, and the similarity measurement threshold is adaptively adjusted so that the adjusted similarity measurement threshold can flexibly respond to the local deformation caused by blood flow changes.

[0012] Based on the adjusted similarity measurement threshold, the multimodal images are finely registered, so that the registration algorithm iteratively optimizes the images according to the dynamically adjusted similarity measurement threshold to ensure spatial alignment and structural matching between images.

[0013] Preferably, the image segmentation technology is used to extract blood flow information from each input retinal blood flow image, and the specific steps are as follows:

[0014] In the process of blood flow information extraction, each preprocessed retinal blood flow image is segmented to separate different areas in the image so as to accurately analyze the blood flow related areas;

[0015] After segmenting the vascular area, the feature extraction method is used to extract the blood flow characteristics, converting the complex image data into numerical information that is easy to analyze, providing data support for the subsequent analysis and prediction of dynamic changes in blood flow.

[0016] Preferably, key features of dynamic changes in blood flow are screened out from the extracted blood flow information, wherein the screened key features include changes in the amplitude of blood flow pulsation and changes in the degree of vascular curvature over time. For each retinal blood flow image, the screened changes in the amplitude of blood flow pulsation and changes in the degree of vascular curvature over time are analyzed in detail, and blood flow pulse fluctuation reference values ​​and vascular curvature change reference values ​​are generated respectively. The degree of change in the dynamic characteristics of blood flow is quantified by the blood flow pulse fluctuation reference values ​​and vascular curvature change reference values.

[0017] Preferably, for each retinal blood flow image, the specific steps of performing a detailed analysis on the amplitude change of the screened blood flow pulsation to generate a blood flow pulse fluctuation reference value are as follows:

[0018] First, the blood flow signal of each frame of retinal blood flow image is carefully extracted by detecting the fluctuation of blood flow signal in time. In order to quantify the change of pulsation amplitude, a blood flow pulse amplitude function is defined , the definition expression of blood flow pulse amplitude function is: ,in, Indicates at location The change in blood flow amplitude at the location reflects the instantaneous change of blood flow at that location. Indicates at location and the blood flow signal value at time t, and Respectively indicated at the position The maximum and minimum values ​​of the blood flow signal at

[0019] After the blood flow pulsation amplitude is extracted, the change of blood flow pulsation amplitude is converted into an overall blood flow pulse fluctuation reference value, which is used to quantify the change of blood flow dynamic characteristics in the entire retinal blood flow image. The blood flow pulse fluctuation reference value can integrate the change of local blood flow pulsation amplitude and reflect it as a global indicator. The calculation expression is: ,in: is the reference value of blood flow pulse fluctuation, Is with regional location The relevant weight reflects the contribution of this location to the overall blood flow dynamics. is the attenuation factor, which is used to control the impact of large amplitude pulsation changes. p is an exponential weighting factor that adjusts the nonlinear contribution of the pulsation amplitude to the overall result.

[0020] Preferably, for each retinal blood flow image, the specific steps of performing a detailed analysis on the changes in the curvature of the screened blood vessels over time to generate a reference value of the blood vessel curvature change are as follows:

[0021] First, for each vascular region in the retinal blood flow image, the instantaneous curvature of the vessel is calculated using the following formula: ,in, Indicates that the blood vessel is at the bend point s The instantaneous curvature at Along the blood vessels s The coordinates of the location, is the tangent slope of the blood vessel, is the rate of change of the curvature of the blood vessel;

[0022] The instantaneous curvature of the blood vessel The curvature change of the entire blood vessel is quantitatively analyzed, and the reference value of the blood vessel curvature change is calculated. The reference value of the blood vessel curvature change quantifies the intensity of the blood flow dynamics by measuring the amplitude of the change of the blood vessel curvature over time. The calculation expression is: ,in, is the reference value of vascular curvature change, represents the spatial rate of change of the vascular curvature, is a smoothing weighting factor for curvature changes, and are the starting point and the ending point of the vascular region, respectively. The integration process is used to accumulate the curvature change intensity of the entire vascular segment.

[0023] Preferably, for each retinal blood flow image, a blood flow pulse fluctuation reference value and a blood vessel curvature change reference value generated by a detailed analysis of the change in the amplitude of blood flow pulsation and the change in the degree of vascular curvature over time are input into a pre-trained support vector machine model, and the blood flow dynamic characteristic intensity is generated by the support vector machine model, and the blood flow dynamic change trend in different regions is predicted based on the blood flow dynamic characteristic intensity.

[0024] Preferably, the similarity measurement threshold is dynamically adjusted according to the blood flow dynamic characteristics predicted by the support vector machine model, and the specific steps are as follows:

[0025] The blood flow dynamic characteristics of each local area are predicted by the support vector machine model, and the blood flow change index of the area is obtained to quantify the dynamic change trend of blood flow. In order to further quantify the deformation characteristics of the local area, the regional deformation factor is introduced. The calculation expression of the regional deformation factor is: ,in: is the regional deformation factor, which reflects the influence of the dynamic changes of blood flow in the local area on the deformation. The blood flow change index generated by the support vector machine model indicates the intensity of the dynamic change of regional blood flow. are parameters for adjusting weights, respectively controlling the nonlinear effect of the blood flow change index on deformation, the dynamic range of deformation characteristics, and the smoothness of exponential growth;

[0026] After calculating the regional deformation factor, the initial similarity measurement threshold is calculated according to the regional deformation factor. The calculation formula of the initial similarity measurement threshold is: ,in: is the initial similarity measurement threshold, which is used to measure the similarity judgment standard during image registration. is the direct adjustment weight of the regional deformation factor on the initial similarity measurement threshold, To control the intensity of the nonlinear effect of the regional deformation factor on the initial similarity metric threshold.

[0027] Preferably, according to the regional deformation factor and the initial similarity measurement threshold, the similarity measurement threshold is further dynamically adjusted to adapt to the local deformation caused by the dynamic change of blood flow. The adaptive adjustment formula is: ,in: is the dynamically adjusted similarity measurement threshold used in the final registration algorithm. is an adjustment factor used to enhance or suppress the effect of regional deformation on the threshold. is a parameter that controls the sensitivity of the deformation factor to the adjustment function. It is the balance point of the deformation factor and determines the turning point of the adjustment function.

[0028] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0029] The present invention uses a support vector machine model to accurately predict the dynamic change trend of blood flow in different regions, and dynamically adjusts the similarity measurement threshold for areas with drastic blood flow changes, so as to flexibly respond to local natural deformations and avoid registration failures caused by dynamic changes such as blood flow fluctuations, vascular expansion or contraction. By introducing a dynamically adjusted threshold standard, the registration algorithm has stronger robustness and adaptability, and more accurate spatial alignment and structural matching are achieved in multimodal images, especially in local areas with significant blood flow changes, which significantly improves the overall accuracy and reliability of registration, and provides more accurate image data support for subsequent retinal disease diagnosis and pathological analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0031] Figure 1 The present invention is a method flow chart of a registration method for multimodal retinal blood flow image processing. DETAILED DESCRIPTION

[0032] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0033] The present invention provides Figure 1 A registration method for multimodal retinal blood flow image processing is shown, comprising the following steps:

[0034] First, retinal blood flow image data from different modalities are received to prepare for subsequent data processing and registration. After receiving retinal blood flow images of different modalities, image preprocessing is performed to reduce the differences between different modalities and ensure that the images are analyzed and registered under the same standard.

[0035] Retinal blood flow image data of different modalities, such as OCT (optical coherence tomography), fluorescein angiography (FA), retinal blood flow imaging (such as laser Doppler flowmeter), etc. These images are obtained by different imaging techniques and reflect different structural and functional information of the retina, but they may differ in spatial resolution, image quality, contrast, etc. The role of this preliminary step is to ensure that the required multimodal data is collected and to prepare for subsequent data processing and alignment. Images of different modalities provide complementary perspectives for the same retinal area. Therefore, integrating these images helps to obtain more comprehensive and accurate retinal blood flow information.

[0036] Common preprocessing techniques include:

[0037] Normalization: Adjust the brightness, contrast and other values ​​of the image to the same range to reduce the impact of differences in imaging technology.

[0038] Denoising: Apply filtering algorithms to remove noise from images to ensure the accuracy of subsequent feature extraction.

[0039] Preliminary registration adjustment: Perform preliminary alignment of images of different modalities to ensure that they roughly overlap in space, laying the foundation for subsequent fine registration.

[0040] The role of image preprocessing is to reduce the inconsistency caused by modality differences, image noise and other factors, so as to make the subsequent feature extraction and registration process more stable and reliable.

[0041] Extract blood flow information from each pre-processed retinal blood flow image to provide a basis for subsequent blood flow dynamic characteristics analysis. Filter out key features of blood flow dynamic changes from the extracted blood flow information, analyze the filtered features in detail, and quantify the degree of change in blood flow dynamic characteristics.

[0042] Using image segmentation technology, blood flow information is extracted from each input retinal blood flow image. The specific steps are as follows:

[0043] In the process of blood flow information extraction, each preprocessed retinal blood flow image is first segmented to separate different areas in the image (such as blood vessels, retinal background, etc.) so as to accurately analyze the blood flow related areas;

[0044] Usually, the vascular region is the key to blood flow information, so the main task of image segmentation is to identify and extract the vascular network. Images can be segmented using methods based on thresholds, edge detection, region growing, or deep learning algorithms (such as U-Net, etc.). The main function of this step is to effectively segment the image and clarify the boundaries between the vascular region and other background regions, providing clear regional divisions for subsequent blood flow information extraction and analysis. Accurate segmentation of vascular regions is the basis for all subsequent blood flow analysis, so high-quality segmentation results are crucial to the entire blood flow information extraction process.

[0045] After segmenting the blood vessel area, the feature extraction method is used to extract the blood flow characteristics, converting the complex image data into numerical information that is easy to analyze, providing data support for the subsequent analysis and prediction of blood flow dynamic changes;

[0046] Blood flow characteristics usually include the morphology of blood vessels (such as diameter, length, curvature, etc.), blood flow velocity (velocity measured by specific imaging techniques such as laser Doppler flowmeters), and directionality of blood flow. These features can reflect the dynamic state of retinal blood flow. Common feature extraction methods include image-based texture analysis, edge detection, morphological analysis, etc., and can be combined with physical model-based calculation methods, such as blood flow simulation, to extract more accurate blood flow-related data. The purpose of blood flow feature extraction is to convert complex image data into meaningful numerical information that can be used for subsequent analysis. By extracting key features of blood vessels (such as diameter, curvature, etc.), the dynamic state of blood flow can be quantified, thereby providing data support for the subsequent analysis and prediction of dynamic changes in blood flow.

[0047] The key features of the dynamic changes of blood flow are screened out from the extracted blood flow information, wherein the screened key features include the changes in the amplitude of blood flow pulsation and the changes in the degree of vascular curvature over time. For each retinal blood flow image, the screened changes in the amplitude of blood flow pulsation and the changes in the degree of vascular curvature over time are analyzed in detail, and blood flow pulse fluctuation reference values ​​and vascular curvature change reference values ​​are generated respectively. The degree of change in the dynamic characteristics of blood flow is quantified by the blood flow pulse fluctuation reference values ​​and the vascular curvature change reference values.

[0048] When the amplitude of blood flow pulsation in the retinal blood flow image changes greatly, it usually indicates that the dynamic change of retinal blood flow is high. The amplitude of blood flow pulsation reflects the degree of fluctuation of blood flow in blood vessels, and this fluctuation is usually closely related to factors such as blood pressure changes, blood flow velocity, and vascular contraction and relaxation. Under normal circumstances, blood flow pulsation will maintain a certain regularity, but when the amplitude of blood flow pulsation suddenly increases, it may be caused by factors such as local blood vessel expansion, contraction, changes in blood viscosity, or changes in the elasticity of the blood vessel wall. This large amplitude of pulsation usually indicates that the blood flow state has undergone more drastic fluctuations or abnormalities, which may be related to retinal hemodynamic changes (such as vascular lesions, retinal ischemia, etc.). Therefore, a large amplitude change in blood flow pulsation is an important sign of strong dynamic changes in blood flow.

[0049] For each retinal blood flow image, the specific steps of performing a detailed analysis of the amplitude changes of the screened blood flow pulsation to generate a blood flow pulse fluctuation reference value are as follows:

[0050] First, the blood flow signal of each frame of retinal blood flow image is carefully extracted by pulse amplitude. This can be achieved by detecting the fluctuation pattern of blood flow signal in time, especially the changes of peaks and troughs of blood flow in local areas. In order to quantify the changes of pulse amplitude, a blood flow pulse amplitude function is defined. , the definition expression of blood flow pulse amplitude function is: ,in, Indicates at location The change in blood flow amplitude at the location reflects the instantaneous change of blood flow at that location. Indicates at location and the blood flow signal value at time t, and Respectively indicated at the position The maximum and minimum values ​​of the blood flow signal at the location where the blood flow is located reflect the amplitude of the peak-to-valley fluctuation of the blood flow on a time scale, thereby capturing the dramatic changes in the dynamics of the blood flow.

[0051] Through this step, the blood flow pulsation amplitude of each pixel or local area can be extracted. This amplitude change directly reflects the dynamic characteristics of blood flow. The amplitude of blood flow pulsation is usually presented in the form of fluctuations in visual images. Therefore, the role of this step is to quantify the intensity of dynamic changes in blood flow through accurate amplitude calculation, laying the foundation for the subsequent generation of pulse fluctuation index.

[0052] After the blood flow pulsation amplitude is extracted, the change of blood flow pulsation amplitude is converted into an overall blood flow pulse fluctuation reference value, which is used to quantify the change of blood flow dynamic characteristics in the entire retinal blood flow image. The blood flow pulse fluctuation reference value can integrate the change of local blood flow pulsation amplitude and reflect it as a global indicator. The calculation expression is: ,in: is the reference value of blood flow pulse fluctuation, Is with regional location The relevant weights can usually be obtained through the density of the vascular network or fluid dynamics simulation, reflecting the contribution of the location to the overall blood flow dynamics. is the attenuation factor, which is used to control the impact of large amplitude pulsation changes. p is an exponential weighting factor used to adjust the nonlinear contribution of the pulsation amplitude to the overall result;

[0053] The purpose of this step is to integrate the local changes in blood flow pulsation amplitude into a global blood flow pulse fluctuation reference value, so that the blood flow pulse fluctuation reference value can accurately reflect the intensity of blood flow dynamics in the entire image. By using exponential weighting and attenuation, the index gives higher weights to larger changes, thereby ensuring that areas with violent blood flow fluctuations contribute more to the final result. This global indicator can not only quantify the dynamic changes of blood flow, but also make comparisons at different time points or between different images, thereby helping to detect and analyze abnormal changes in retinal blood flow.

[0054] It can be seen from the blood flow pulse fluctuation reference value that for each retinal blood flow image, the larger the performance value of the blood flow pulse fluctuation reference value generated by a detailed analysis of the amplitude change of the screened blood flow pulsation, the higher the degree of dynamic change of the blood flow in the retinal blood flow image. The blood flow pulse fluctuation reference value reflects the intensity of blood flow fluctuation in time and space by quantifying the amplitude change of local blood flow pulsation. When the blood flow pulse fluctuation reference value is large, it means that the blood flow has experienced drastic pulsation changes in the image, which may be related to the expansion and contraction of blood vessels, or the sharp change of blood flow velocity, and usually means that the dynamic characteristics of blood flow have fluctuated significantly or abnormally. On the contrary, when the blood flow pulse fluctuation reference value is small, it means that the blood flow in this area changes relatively smoothly and the dynamic changes are small, indicating that the dynamic characteristics of blood flow in this image are relatively stable. Therefore, the blood flow pulse fluctuation reference value provides an effective quantitative indicator for describing the dynamic change intensity of retinal blood flow, and the high or low blood flow pulse fluctuation reference value directly reflects the drastic degree of dynamic change of blood flow.

[0055] When the curvature of blood vessels in retinal blood flow images changes greatly over time, it usually indicates that the dynamic change of retinal blood flow is high. This is because the curvature of blood vessels is closely related to the dynamic characteristics of blood flow. Under the influence of hemodynamic changes, blood vessels may expand, contract or change shape. When blood flow, blood flow velocity or blood pressure changes significantly, the morphology of blood vessels will change accordingly, especially in the curved parts of local blood vessels. For example, in the state of hypertension or hypotension, blood vessels may be more curved or deformed due to the sharp fluctuations in blood flow. In addition, changes in the curvature of blood vessels may also reflect the pressure difference or flow resistance change in different areas of blood flow. Therefore, by observing the changes in the curvature of blood vessels, the dynamic change intensity of retinal blood flow can be effectively revealed.

[0056] For each retinal blood flow image, the specific steps for analyzing the changes in the curvature of the screened blood vessels over time and generating reference values ​​for the changes in blood vessel curvature are as follows:

[0057] First, for each vascular region in the retinal blood flow image, the instantaneous curvature of the vessel is calculated. The curvature of the vessel is a parameter that describes the degree of curvature of the vessel shape and is calculated using the following formula: ,in, Indicates that the blood vessel is at the bend point s The instantaneous curvature at Along the blood vessels s The coordinates of the location, is the tangent slope of the blood vessel, is the rate of change of the curvature of the blood vessel;

[0058] The purpose of this step is to calculate the instantaneous change of the vessel curvature based on the vessel geometry. It can reflect the strength of local curvature of blood vessels and is an important indicator for capturing the dynamic changes of blood flow. It can reveal the instantaneous dynamic characteristics of blood flow, especially when the curvature of the blood vessels is large. By calculating the curvature at each point along the blood vessels, the changing trend of the blood vessels can be accurately captured, especially when the blood flow fluctuates violently, the change of the blood vessel curvature will also increase accordingly.

[0059] The instantaneous curvature of the blood vessel The curvature change of the entire blood vessel is quantitatively analyzed, and the reference value of the blood vessel curvature change is calculated. The reference value of the blood vessel curvature change quantifies the intensity of the blood flow dynamics by measuring the amplitude of the change of the blood vessel curvature over time. The calculation expression is: ,in, is the reference value of vascular curvature change, represents the spatial rate of change of the vessel curvature, is a smoothing weighting factor for curvature changes, and are the starting point and the ending point of the vascular region, respectively. The integration process is used to accumulate the intensity of the curvature change of the entire vascular segment;

[0060] This step quantifies the severity of the change in vascular morphology over time by integrating the rate of change of vascular curvature. The magnitude of the curvature change rate can reveal whether the blood vessel has undergone a large deformation, and the weighting factor A higher weight is given to the area with larger vascular curvature, so as to more accurately reflect the intensity of the dynamic change of blood flow. Ultimately, the reference value of vascular curvature change can be used to characterize the intensity of the dynamic characteristic change of the entire vascular area, thereby providing a quantitative basis for the evaluation of the dynamic change of blood flow.

[0061] It can be seen from the reference value of vascular curvature change that for each retinal blood flow image, the larger the performance value of the vascular curvature change reference value generated by a detailed analysis of the change in the degree of curvature of the screened blood vessels over time, the more dramatic the change in the degree of curvature of the blood vessels in the retinal blood flow image over time, thereby reflecting the higher degree of dynamic change of retinal blood flow. The curvature of the blood vessels is closely related to the change of blood flow. When the blood flow fluctuates violently, the vascular morphology will change significantly accordingly, especially the curvature of the blood vessels in the local area will increase. Therefore, the reference value of vascular curvature change can quantify the change in the dynamic characteristics of blood flow. The larger the value, the more significant the change in vascular morphology, the more active the blood flow dynamics, and there may be stronger blood flow fluctuations or pressure changes; conversely, when the vascular curvature changes slightly or stably, it means that the dynamic change of blood flow is small, the vascular morphology remains stable, and the blood flow state is relatively stable.

[0062] The analyzed and quantified blood flow dynamic characteristics are input into the pre-trained support vector machine model to predict the blood flow dynamic change trend in different areas;

[0063] For each retinal blood flow image, a detailed analysis of the changes in the amplitude of blood flow pulsation and the changes in the degree of vascular curvature over time is performed to generate blood flow pulse fluctuation reference values ​​and vascular curvature change reference values, which are input into a pre-trained support vector machine model. The blood flow dynamic characteristic intensity is generated by the support vector machine model, and the blood flow dynamic change trend in different regions is predicted based on the blood flow dynamic characteristic intensity.

[0064] A pre-trained support vector machine (SVM) model refers to a machine learning model that has been trained and optimized after inputting specific blood flow pulse fluctuation reference values ​​and vascular curvature change reference values. It can automatically identify the dynamic change pattern of blood flow and make predictions. SVM is a supervised learning model that is mainly used for classification and regression problems. It separates data points of different categories by constructing hyperplanes in the feature space. In multimodal retinal blood flow image analysis, SVM is trained to identify and predict the trend of dynamic changes in blood flow. Specifically, it learns the two features of blood flow pulse fluctuation reference value (reflecting the change in the amplitude of blood flow fluctuation) and vascular curvature change reference value (reflecting the change in the degree of vascular curvature over time) to build a model that can effectively identify the intensity of dynamic changes in blood flow.

[0065] During the training process, SVM will use a large amount of labeled training data (i.e., image data with known dynamic changes in blood flow) to adjust the model parameters according to the blood flow characteristics of different regions, and finally determine a hyperplane that can accurately predict unknown data. The support vector machine ensures the optimality of classification by maximizing the interval (i.e., the distance between different categories), thereby achieving efficient prediction of dynamic changes in blood flow. In practical applications, the pre-trained SVM model can not only process multi-dimensional features such as blood flow pulse fluctuation reference values ​​and vascular curvature change reference values, but also process different types of retinal blood flow images, making the model more universal and flexible.

[0066] In actual application, the trained SVM model will generate a blood flow dynamic characteristic intensity based on the input blood flow pulse fluctuation reference value and vascular curvature change reference value. This blood flow dynamic characteristic intensity combines the intensity of blood flow pulsation and the change of vascular morphology in the image, so as to predict the blood flow dynamic trend in the area. Areas with higher blood flow dynamic characteristic intensity usually indicate drastic changes in blood flow dynamics, and there may be phenomena such as increased blood flow, blood pressure fluctuations, and changes in vascular morphology. Areas with lower blood flow dynamic characteristic intensity indicate that blood flow dynamics are relatively stable, local vascular morphology does not change much, and the blood flow state is in a relatively stable state. By analyzing the blood flow pulse fluctuations and vascular curvature characteristics in different regions, the SVM model can effectively identify areas where blood flow has changed significantly, especially when some local blood vessels have changed dramatically.

[0067] The support vector machine model is not specifically limited here, and can realize the reference value of blood flow pulse fluctuation and reference values ​​of vascular curvature changes Comprehensive analysis to generate blood flow dynamic characteristics intensity The support vector machine model can be used. In order to realize the technical solution of the present invention, the present invention provides a specific implementation method; the intensity of the dynamic characteristics of blood flow The generated expression is: , where , Blood flow pulse fluctuation reference value and reference values ​​of vascular curvature changes The preset scaling factor of , All are greater than 0. Preset proportional coefficient ( , ) is the weight setting for the blood flow pulse fluctuation reference value and the vascular curvature change reference value in the process of generating the blood flow dynamic characteristic intensity. Its function is to adjust the relative contribution ratio of these two features in the formula. Since the influence of blood flow pulse fluctuation and vascular curvature change on blood flow dynamics may be different in actual situations, by introducing a preset proportional coefficient, the importance of the two can be distinguished and weighted in the model. For example, if the blood flow pulse fluctuation has a more significant impact on blood flow changes in a certain situation, it can be increased by to increase its weight in the formula; similarly, if the change in vascular curvature is more critical, The preset proportionality coefficient is usually set based on experience, experimental data or the requirements of a specific application scenario, and its purpose is to improve the generation accuracy of the intensity of the blood flow dynamic characteristics so that it is more in line with the law of actual blood flow dynamic changes.

[0068] It can be seen from the intensity of the blood flow dynamic characteristics that, for each retinal blood flow image, the greater the expression value of the blood flow pulse fluctuation reference value generated by a detailed analysis of the amplitude change of the screened blood flow pulsation, the greater the expression value of the vascular curvature change reference value generated by a detailed analysis of the change in the screened blood vessel curvature over time, that is, the greater the expression value of the blood flow dynamic characteristic intensity generated when the blood flow dynamic change trend in the retinal blood flow image is predicted by the pre-trained support vector machine model, the higher the degree of dynamic change of the retinal blood flow in the area, and vice versa.

[0069] The similarity measurement threshold is dynamically adjusted according to the dynamic characteristics of blood flow predicted by the support vector machine model. For areas with drastic blood flow changes, the deformation degree of the area is judged by the prediction results, and the similarity measurement threshold is adaptively adjusted so that the adjusted similarity measurement threshold can flexibly respond to the local deformation caused by blood flow changes.

[0070] Flexible thresholds for deformation regions: In regions with large blood flow variations, the similarity metric threshold should be appropriately relaxed to allow for larger local deformations.

[0071] Strict thresholds in stable areas: In areas where blood flow changes are small or relatively stable, the similarity metric threshold can be set more strictly to ensure the accuracy of the registration results.

[0072] By dynamically adjusting the similarity measurement threshold, the registration strategy can be finely adjusted according to the dynamic characteristics of blood flow, effectively coping with local deformation caused by blood flow changes, thereby avoiding mismatching caused by blood flow changes and improving registration accuracy.

[0073] The similarity measurement threshold is dynamically adjusted according to the blood flow dynamic characteristics predicted by the support vector machine model. The specific steps are as follows:

[0074] The blood flow dynamic characteristics of each local area are predicted by the support vector machine model, and the blood flow change index of the area is obtained. The blood flow change index combines the information of the blood flow pulse fluctuation reference value and the vascular curvature change reference value to quantify the dynamic change trend of blood flow. In order to further quantify the deformation characteristics of the local area, the regional deformation factor is introduced. The calculation expression of the regional deformation factor is: ,in: is the regional deformation factor, which reflects the influence of the dynamic changes of blood flow in the local area on the deformation. The blood flow change index generated by the support vector machine model indicates the intensity of the dynamic change of regional blood flow. are parameters for adjusting weights, respectively controlling the nonlinear effect of the blood flow change index on deformation, the dynamic range of deformation characteristics, and the smoothness of exponential growth;

[0075] By calculating the regional deformation factor, the dynamic characteristics of blood flow are further converted into deformation parameters that can be directly used for registration analysis. The size of the deformation factor directly reflects the intensity of the dynamic change of the region, providing a basis for the dynamic adjustment of the subsequent similarity measurement threshold.

[0076] After calculating the regional deformation factor, the initial similarity measurement threshold is calculated according to the regional deformation factor. The calculation formula of the initial similarity measurement threshold is: ,in: is the initial similarity measurement threshold, which is used to measure the similarity judgment standard during image registration. is the direct adjustment weight of the regional deformation factor on the initial similarity measurement threshold, To control the intensity of the nonlinear effect of the regional deformation factor on the initial similarity metric threshold;

[0077] During the dynamic adjustment process, the blood flow changes drastically in the area ( Larger), the initial similarity measurement threshold ( ) will become smaller, thus allowing a larger local deformation range and flexibly responding to local deformation caused by blood flow changes; while for areas with relatively stable blood flow changes ( Smaller), the initial similarity measurement threshold ( ) will tend to a higher value to ensure the accuracy of the registration result. The initial similarity metric threshold ( ) provides a flexible matching criterion for the subsequent registration algorithm, so that the dynamic changes of blood flow in different regions can be accurately processed, thereby improving the robustness of the registration.

[0078] According to the regional deformation factor and the initial similarity measurement threshold, the similarity measurement threshold is further dynamically adjusted to adapt to the local deformation caused by the dynamic changes of blood flow. The adaptive adjustment formula is: ,in: is the dynamically adjusted similarity measurement threshold used in the final registration algorithm. is an adjustment factor used to enhance or suppress the effect of regional deformation on the threshold. is a parameter that controls the sensitivity of the deformation factor to the adjustment function. is the balance point of the deformation factor, which determines the turning point of the adjustment function;

[0079] When the regional deformation factor exceeds the set balance point, the similarity metric threshold will be significantly reduced to allow a larger local deformation range to adapt to the dramatic dynamic changes in blood flow; conversely, when the deformation factor is small, the threshold will be maintained at a high level to ensure the accuracy of registration. Through the adaptively adjusted similarity metric threshold, the registration algorithm can flexibly adjust the matching criteria according to the severity of blood flow changes. This step further optimizes the threshold so that it can handle both areas with dramatic dynamic changes and areas with smaller changes accurately, ultimately significantly improving the robustness and accuracy of multimodal retinal blood flow image registration.

[0080] Based on the adjusted similarity measurement threshold, the multimodal images are finely registered, so that the registration algorithm iteratively optimizes the images according to the dynamically adjusted similarity measurement threshold to ensure spatial alignment and structural matching between images;

[0081] The step of fine registration based on the adjusted similarity metric threshold means that after the similarity matching criteria of the blood flow dynamic change area are dynamically adjusted, the registration algorithm gradually optimizes and iteratively corrects the multimodal images, so that the images can be accurately aligned in spatial position and local structure. Through the dynamically adjusted similarity metric threshold, the registration algorithm can flexibly cope with deformations in different areas, especially in areas with drastic changes in blood flow. The algorithm allows larger deformation matching errors, while maintaining stricter matching standards in areas with stable changes. The fine registration process gradually reduces the residual between images through iterative optimization, maximizes the structural similarity of the two images, and ensures accurate spatial alignment and structural matching in areas with dynamic changes in blood flow and non-dynamic changes. This process can effectively avoid mismatching, improve the overall accuracy and robustness of multimodal image registration, and thus provide more reliable data support for subsequent clinical analysis and diagnosis. In this way, the registration algorithm can flexibly cope with dynamic changes in blood flow and local deformation, and maximize the accuracy of image registration.

[0082] The above-mentioned registration method for multimodal retinal blood flow image processing can effectively solve the problems of mismatching and decreased registration accuracy caused by fixed similarity thresholds in traditional registration technology, and realize adaptive similarity measurement threshold adjustment based on the dynamic characteristics of blood flow. This method accurately predicts the dynamic change trend of blood flow in different regions through the support vector machine model, and dynamically adjusts the similarity measurement threshold for areas with drastic blood flow changes, so as to flexibly respond to local natural deformations and avoid registration failures caused by dynamic changes such as blood flow fluctuations, vascular expansion or contraction. By introducing a dynamically adjusted threshold standard, the registration algorithm has stronger robustness and adaptability, and more accurate spatial alignment and structural matching are achieved in multimodal images, especially in local areas with significant blood flow changes, which significantly improves the overall accuracy and reliability of registration, and provides more accurate image data support for subsequent retinal disease diagnosis and pathological analysis.

[0083] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0084] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0085] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A registration method for multimodal retinal blood flow image processing, characterized in that: The following steps are involved: First, retinal blood flow image data from different modalities are received to prepare for subsequent data processing and registration. After receiving retinal blood flow images of different modalities, image preprocessing is performed to reduce the differences between different modalities and ensure that the images are analyzed and registered under the same standard. Extract blood flow information from each pre-processed retinal blood flow image to provide a basis for subsequent blood flow dynamic characteristics analysis. Filter out key features of blood flow dynamic changes from the extracted blood flow information, analyze the filtered features in detail, and quantify the degree of change in blood flow dynamic characteristics. The analyzed and quantified blood flow dynamic characteristics are input into the pre-trained support vector machine model to predict the blood flow dynamic change trend in different areas; The similarity measurement threshold is dynamically adjusted according to the dynamic characteristics of blood flow predicted by the support vector machine model. For areas with drastic blood flow changes, the deformation degree of the area is judged by the prediction results, and the similarity measurement threshold is adaptively adjusted so that the adjusted similarity measurement threshold can flexibly respond to the local deformation caused by blood flow changes. Based on the adjusted similarity measurement threshold, the multimodal images are finely registered, so that the registration algorithm iteratively optimizes the images according to the dynamically adjusted similarity measurement threshold to ensure spatial alignment and structural matching between images.

2. A registration method for multimodal retinal blood flow image processing according to claim 1, characterized in that: Using image segmentation technology, blood flow information is extracted from each input retinal blood flow image. The specific steps are as follows: In the process of blood flow information extraction, each preprocessed retinal blood flow image is segmented to separate different areas in the image so as to accurately analyze the blood flow related areas; After segmenting the vascular area, the feature extraction method is used to extract the blood flow characteristics, converting the complex image data into numerical information that is easy to analyze, providing data support for the subsequent analysis and prediction of dynamic changes in blood flow.

3. The registration method for multimodal retinal blood flow image processing according to claim 1, characterized in that: The key features of the dynamic changes of blood flow are screened out from the extracted blood flow information, wherein the screened key features include the changes in the amplitude of blood flow pulsation and the changes in the degree of vascular curvature over time. For each retinal blood flow image, the screened changes in the amplitude of blood flow pulsation and the changes in the degree of vascular curvature over time are analyzed in detail, and blood flow pulse fluctuation reference values ​​and vascular curvature change reference values ​​are generated respectively. The degree of change in the dynamic characteristics of blood flow is quantified by the blood flow pulse fluctuation reference values ​​and the vascular curvature change reference values.

4. A registration method for multimodal retinal blood flow image processing according to claim 3, characterized in that: For each retinal blood flow image, the specific steps of performing a detailed analysis of the amplitude changes of the screened blood flow pulsation to generate a blood flow pulse fluctuation reference value are as follows: First, the blood flow signal of each frame of retinal blood flow image is carefully extracted by detecting the fluctuation of blood flow signal in time. In order to quantify the change of pulsation amplitude, a blood flow pulse amplitude function is defined , the definition expression of blood flow pulse amplitude function is: ,in, Indicates at location The change in blood flow amplitude at the location reflects the instantaneous change of blood flow at that location. Indicates at location and the blood flow signal value at time t, and Respectively indicated at the position The maximum and minimum values ​​of the blood flow signal at After the blood flow pulsation amplitude is extracted, the change of blood flow pulsation amplitude is converted into an overall blood flow pulse fluctuation reference value, which is used to quantify the change of blood flow dynamic characteristics in the entire retinal blood flow image. The blood flow pulse fluctuation reference value can integrate the change of local blood flow pulsation amplitude and reflect it as a global indicator. The calculation expression is: ,in: is the reference value of blood flow pulse fluctuation, Is with regional location The relevant weight reflects the contribution of this location to the overall blood flow dynamics. is the attenuation factor, which is used to control the impact of large amplitude pulsation changes. p is an exponential weighting factor that adjusts the nonlinear contribution of the pulsation amplitude to the overall result.

5. The registration method for multimodal retinal blood flow image processing according to claim 3, characterized in that: For each retinal blood flow image, the specific steps for analyzing the changes in the curvature of the screened blood vessels over time and generating reference values ​​for the changes in blood vessel curvature are as follows: First, for each vascular region in the retinal blood flow image, the instantaneous curvature of the vessel is calculated using the following formula: ,in, Indicates that the blood vessel is at the bend point s The instantaneous curvature at Along the blood vessels s The coordinates of the location, is the tangent slope of the blood vessel, is the rate of change of the curvature of the blood vessel; The instantaneous curvature of the blood vessel The curvature change of the entire blood vessel is quantitatively analyzed, and the reference value of the blood vessel curvature change is calculated. The reference value of the blood vessel curvature change quantifies the intensity of the blood flow dynamics by measuring the amplitude of the change of the blood vessel curvature over time. The calculation expression is: ,in, is the reference value of vascular curvature change, represents the spatial rate of change of the vascular curvature, is a smoothing weighting factor for curvature changes, and are the starting point and the ending point of the vascular region, respectively. The integration process is used to accumulate the curvature change intensity of the entire vascular segment.

6. The registration method for multimodal retinal blood flow image processing according to claim 3, characterized in that: For each retinal blood flow image, a detailed analysis of the changes in the amplitude of blood flow pulsation and the changes in the degree of vascular curvature over time is performed to generate blood flow pulse fluctuation reference values ​​and vascular curvature change reference values, which are input into a pre-trained support vector machine model. The blood flow dynamic characteristic intensity is generated by the support vector machine model, and the blood flow dynamic change trend in different regions is predicted based on the blood flow dynamic characteristic intensity.

7. A registration method for multimodal retinal blood flow image processing according to claim 6, characterized in that: The similarity measurement threshold is dynamically adjusted according to the blood flow dynamic characteristics predicted by the support vector machine model. The specific steps are as follows: The blood flow dynamic characteristics of each local area are predicted by the support vector machine model, and the blood flow change index of the area is obtained to quantify the dynamic change trend of blood flow. In order to further quantify the deformation characteristics of the local area, the regional deformation factor is introduced. The calculation expression of the regional deformation factor is: ,in: is the regional deformation factor, which reflects the influence of the dynamic changes of blood flow in the local area on the deformation. The blood flow change index generated by the support vector machine model indicates the intensity of the dynamic change of regional blood flow. are parameters for adjusting weights, respectively controlling the nonlinear effect of the blood flow change index on deformation, the dynamic range of deformation characteristics, and the smoothness of exponential growth; After calculating the regional deformation factor, the initial similarity measurement threshold is calculated according to the regional deformation factor. The calculation formula of the initial similarity measurement threshold is: ,in: is the initial similarity measurement threshold, which is used to measure the similarity judgment standard during image registration. is the direct adjustment weight of the regional deformation factor on the initial similarity measurement threshold, To control the intensity of the nonlinear effect of the regional deformation factor on the initial similarity metric threshold.

8. The registration method for multimodal retinal blood flow image processing according to claim 7, characterized in that: According to the regional deformation factor and the initial similarity measurement threshold, the similarity measurement threshold is further dynamically adjusted to adapt to the local deformation caused by the dynamic changes of blood flow. The adaptive adjustment formula is: ,in: is the dynamically adjusted similarity measurement threshold used in the final registration algorithm. is an adjustment factor used to enhance or suppress the effect of regional deformation on the threshold. is a parameter that controls the sensitivity of the deformation factor to the adjustment function. It is the balance point of the deformation factor and determines the turning point of the adjustment function.

Images