A Method for Suppressing Axial Artifacts in 3D OCTA Based on Morphological Features

Abstract

This invention discloses a morphological feature-based method for axial artifact suppression in three-dimensional OCTA (Optical Coherence Tomography) imaging. The method involves distinguishing between large and small blood vessel regions using an image projection network, and restricting the axial connectivity of the blood vessels based on calculated lateral thickness and polar angle. Finally, Hessian filtering is used to further enhance the restricted image, achieving high-quality, axial artifact-free three-dimensional optical coherence tomography (OCT) vascular imaging of the inner retina. This invention significantly outperforms traditional mean subtraction and vessel enhancement methods in removing axial artifacts, and requires no additional imaging techniques, allowing direct application to existing data.

Description

Technical Field

[0001] This invention relates to the image processing applications of optical coherence tomography (OCTA) of the inner retina, specifically to a method for suppressing axial artifacts in three-dimensional OCTA based on morphological features. Background Technology

[0002] Optical coherence tomography (OCT) is a rapid, label-free imaging tool widely used in clinical ophthalmology. Optical coherence tomography angiography (OCTA) utilizes the dynamic scattering of moving blood cells within vessels to visualize them. OCTA's depth resolution and non-invasive nature allow for repeated examination of the retinal capillary plexus and choroidal capillary system. It can be used to identify characteristic features of retinal diseases, such as retinal microvascular abnormalities and neovascularization caused by diabetic retinopathy. Furthermore, OCTA can quantify subtle microvascular changes within the retinal capillary network. However, the presence of axial tail artifacts caused by multiple scattering makes it difficult to fully utilize the three-dimensional (3D) characteristics of OCTA.

[0003] Axial artifacts are a major drawback of OCTA, causing spatial repetition of certain vascular features and limiting the accurate representation of 3D vascular data. Furthermore, these axial artifacts can disrupt the rod-like structure of vessels, rendering traditional vascular enhancement techniques (such as Hessian-filtering) ineffective. Therefore, proposing a reliable method to suppress axial artifacts in OCTA and restore the three-dimensional features of vessels is crucial for the quantitative characterization of vessels in 3D OCTA.

[0004] The existing technology has the following technical problems:

[0005] 1) The presence of axial artifacts disrupts the original rod-like structure of blood vessels, rendering traditional techniques for enhancing blood vessels based on the rod-like structure ineffective.

[0006] 2) When the length of the axial connected domain of a blood vessel is limited by the transverse thickness of the blood vessel, the presence of background noise will cause two adjacent blood vessel signals under the same A-Scan to merge into a single connected domain, which will lead to the disappearance of deep blood vessels after limiting the axial connected domain.

[0007] 3) Limiting the length of the axial connectivity domain of a blood vessel by using the transverse thickness of the vessel to remove axial artifacts is no longer accurate for vessels with axial flux. Summary of the Invention

[0008] To address the current shortcomings in this field, this invention proposes a 3D OCTA axial artifact suppression method based on morphological features. Specifically, it's a 3D optical coherence tomography (OCTA) axial artifact suppression method based on vessel thickness and vessel enhancement algorithms, and a 3D OCTA axial artifact suppression method based on vessel lateral thickness and vessel enhancement algorithms. This method roughly distinguishes between small and large vessel regions using an image projection network (IPN), calculates the thickness morphological features of vessels in each region, and limits the size of the axial connected region of the vessel signal based on the calculated results, thus achieving initial suppression of axial artifacts. Finally, the merged image with initially suppressed artifacts is processed using a vessel enhancement algorithm to obtain the final 3D OCTA image with axial artifacts removed.

[0009] This invention is achieved through the following technical solution:

[0010] This invention discloses a method for suppressing axial artifacts in three-dimensional OCTA based on morphological features, characterized by comprising the following steps:

[0011] 1) The small and large blood vessels in the 3D OCTA image projection are segmented by the Image Projection Network (IPN) to obtain the segmentation results of the small and large blood vessels in the original image projection; the small blood vessels are capillaries, and the large blood vessels are veins and arteries.

[0012] 2) Multiply the small blood vessel segmentation results and large blood vessel segmentation results of the original image projection with each layer of the 3D image to obtain coarsely segmented small blood vessel region images and large blood vessel region images.

[0013] 3) Use 5%-40% of the average intensity of the small blood vessel region image obtained in 2) as the intensity threshold of the large blood vessel region obtained in 2), remove signals with intensity below the threshold in the large blood vessel region, and obtain the large blood vessel region image with background noise removed.

[0014] 4) Calculate the lateral thickness and polar angle of the blood vessels in the small blood vessel region image obtained in 2) and the large blood vessel region image obtained in 3) after removing background noise, respectively. Then process the lateral thickness value according to the calculated polar angle to obtain the processed lateral thickness values ​​of the small blood vessel region and the large blood vessel region.

[0015] 5) Based on the transverse thickness values ​​of the small blood vessel region and the large blood vessel region obtained in 4), the length of the axial connected region of each blood vessel signal in the small blood vessel region image obtained in 2) and the large blood vessel region image obtained in 3) after removing background noise is restricted respectively. The size of each connected region is restricted to the corresponding transverse thickness after processing, so as to obtain the small blood vessel region image and the large blood vessel region image after removing axial artifacts.

[0016] 6) Combine the small vessel region image obtained in 5) with the large vessel region image after preliminary removal of axial artifacts into a single 3D OCTA image after preliminary removal of axial artifacts;

[0017] 7) The 3D OCTA image obtained in 6) with preliminary removal of axial artifacts is further enhanced with Hessian-filtering to obtain the final 3D OCTA image with axial artifacts removed.

[0018] As a further improvement, in step 3) of this invention, the specific threshold selection method is as follows: calculate the ratio of the average intensity of multiple background noise regions in the large blood vessel region to the average intensity of the small blood vessel region, and determine the maximum value among all ratios. Select the product of a percentage slightly greater than this maximum value and the average intensity of the small blood vessel region as the intensity threshold of the large blood vessel region.

[0019] As a further improvement, in step 4) of this invention, the transverse thickness is processed as follows: for vessels that are not fully axial, the processed transverse thickness value TM is the transverse thickness T calculated by the thickness algorithm divided by the polar angle. The sine value and rounded down:

[0020]

[0021] As a further improvement, in step 5) of the present invention, the method for limiting the length of the axial connected domain of the blood vessel is as follows: taking the maximum intensity value of the axial connected domain of the blood vessel as the center, and limiting the number of voxels of the connected domains above and below the center to half of the transverse thickness value of the blood vessel after processing, TM / 2.

[0022] The beneficial effects of this invention are as follows:

[0023] This method uses processed lateral vessel thickness to limit its axial connectivity, supplemented by Hessian-filtering, to achieve 3D optical coherence tomography (OCT) vascular imaging with axial artifact removal. This invention significantly outperforms traditional mean subtraction and vessel enhancement methods in removing axial artifacts, and requires no additional imaging techniques, allowing direct application to existing 3D OCTA data.

[0024] Since blood vessels with axial artifacts are no longer rod-shaped structures, the processing effect of traditional vascular enhancement methods is not ideal. In steps 4) and 5) of this invention, by calculating the lateral thickness and polar angle of the blood vessel, the axial connected domain of the blood vessel is restricted before vascular enhancement, restoring the blood vessel to an approximate rod-shaped structure. Then, the vascular enhancement algorithm is used for further enhancement, achieving effective removal of axial artifacts.

[0025] In steps 1), 2), and 3) of the present invention, before limiting the length of the axial connected domain of the blood vessel by the transverse thickness of the blood vessel, the original image is divided into two regions, large and small blood vessels, by using IPN, and 5%-40% of the average intensity of the small blood vessel region is used as the intensity threshold of the large blood vessel region to suppress the background noise of the large blood vessel region, thereby avoiding the problem of deep blood vessel signal loss after limiting the axial connected domain.

[0026] For vessels that are not fully axial, the present invention uses its transverse thickness divided by its polar angle in step 4). The sine value is used as the final axial connectivity limit length, thereby accurately reconstructing the structure of blood vessels with axial flux. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the process of the present invention;

[0028] Figure 2 This is a schematic diagram illustrating the calculation of the transverse thickness of blood vessels in this invention;

[0029] Figure 3 This is a schematic diagram illustrating the calculation of the vessel polar angle and the processing of the lateral thickness in this invention.

[0030] Figure 4 This diagram illustrates a comparison of the effects of the present invention with existing methods. Detailed Implementation

[0031] To describe the present invention in more detail, the technical methods of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] This invention relates to a method for suppressing axial artifacts in three-dimensional OCTA based on morphological features. The specific steps are as follows: Figure 1 As shown.

[0033] Taking the acquired 3D OCTA image of the inner retina as an example, the projected image is first segmented into large and small blood vessels using the publicly available IPN network. Large blood vessels refer to arteries and veins, while small blood vessels refer to capillaries. The parameters used for training the IPN network are publicly available default parameters, and the labeled data used to train the model comes from the publicly available OCTA-500 dataset.

[0034] After obtaining the segmentation results of large and small blood vessels in the projected image, they are multiplied by each layer of the 3D OCTA image to obtain two coarsely segmented regions containing large and small blood vessels. Ten different background noise regions within the large blood vessel region are manually selected, and the average intensity of the corresponding regions is calculated. The maximum value of the ratio of the average intensity of each region to the average intensity of the small blood vessel region is determined. A value slightly greater than this maximum value, i.e., 10% of the average intensity of the small blood vessel region, is selected as the intensity threshold for the large blood vessel region, and signals with intensity below the threshold in the large blood vessel region are removed.

[0035] Next, the lateral thickness of the blood vessels in the large and small vessel regions is calculated separately, using the following method: Figure 2 As shown, the OCTA image is first binarized, with voxels of value 1 representing blood vessel signals and voxels of value 0 representing the background. For each blood vessel voxel, the shortest distance to the background is calculated and distance propagation is performed. The maximum distance value within the calculation window centered on the voxel is assigned to the voxel to ensure that the distance values ​​of voxels perpendicular to the blood vessel within the same cross-section are approximately uniform. Subsequently, a correlation operator is used to smooth the distance propagation results, simulating the continuous change in blood vessel thickness, to obtain the final calculated result of the transverse thickness of the blood vessel.

[0036] Before applying axially connected domain constraints based on lateral thickness to blood vessels with axial flux, the lateral thickness needs to be processed. For this, a weighted vector summation method is used to calculate the polar angle of the blood vessel in the 3D OCTA image. Polar angle Defined as the angle between the blood vessel and the -Z axis, ranging from 0° to 90°, such as... Figure 3 As shown. 0° indicates that the blood vessels are distributed entirely along the axial direction, while 90° indicates that the blood vessels are distributed entirely along the horizontal direction. Considering direct calculation... To address the complexity, two auxiliary azimuth angles, β and γ, were introduced to facilitate... The calculations are as follows. The relationships between these angles are as follows:

[0037]

[0038] Where β is defined as the angle between the projection of the blood vessel onto the ZX plane and the X-axis, and γ is defined as the angle between the projection of the blood vessel onto the YZ plane and the -Y axis. Figure 3 ).

[0039] To calculate angles β and γ, a computational window of size 7×7×7 voxels (depending on the vessel size) is first generated, centered on the target voxel on the blood vessel. Within this window, all vectors that may pass through the target voxel are defined, and these vectors are weighted according to their lengths and the intensity variations between adjacent voxels. The β or γ angle of the target voxel is then determined by summing all weighted vectors, and the polar angle of the voxel is derived using the relationship mentioned above. Calculate the polar angle Then, the transverse thickness of the incompletely axially oriented blood vessel was processed using the following equation:

[0040]

[0041] Where T represents the transverse thickness calculated using a thickness algorithm, and TM represents the thickness calculated based on... The thickness after processing.

[0042] By restricting the axial connectivity of blood vessels in the large and small vessel regions to their corresponding transmembrane areas (TMs), we can obtain the large and small vessel regions after preliminary removal of axial artifacts. Then, the large and small vessel regions are merged, and vascular enhancement is performed using Hessian-filtering to obtain the final axial artifact-free 3DOCTA image of the inner retina. The effectiveness of this invention compared to existing artifact removal methods is as follows: Figure 4 As shown, PSNR is the peak signal-to-noise ratio, SSIM is the structural similarity, and MSE is the mean square error. All three evaluation metrics are calculated using adjacent B-scan images. The Hessian-filtering response function used in this invention is:

[0043]

[0044] Where λ1 to λ3 are the three eigenvalues ​​of the Hessian matrix arranged in ascending order of absolute value, and λ ρ The calculation is performed independently at each scale σ to ensure that the enhancement function is normalized at each scale. Its expression is:

[0045]

[0046] The above description of the examples is provided to enable those skilled in the art to understand and apply the present invention. It will be apparent to those skilled in the art that various modifications can be made to the above examples, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A method for suppressing axial artifacts in three-dimensional OCTA based on morphological features, characterized in that, Includes the following steps: 1) The small and large blood vessels in the 3D OCTA image projection are segmented by the Image Projection Network (IPN) to obtain the segmentation results of the small and large blood vessels in the original image projection; the small blood vessels are capillaries, and the large blood vessels are veins and arteries. 2) Multiply the small blood vessel segmentation results and large blood vessel segmentation results of the original image projection with each layer of the 3D image to obtain coarsely segmented small blood vessel region images and large blood vessel region images. 3) Use 5%-40% of the average intensity of the small blood vessel region image obtained in 2) as the intensity threshold of the large blood vessel region obtained in 2), remove signals with intensity below the threshold in the large blood vessel region, and obtain the large blood vessel region image with background noise removed. 4) Calculate the lateral thickness and polar angle of the blood vessels in the small blood vessel region image obtained in 2) and the large blood vessel region image obtained in 3) after removing background noise, respectively. Then process the lateral thickness value according to the calculated polar angle to obtain the processed lateral thickness values ​​of the small blood vessel region and the large blood vessel region. 5) Based on the transverse thickness values ​​of the small blood vessel region and the large blood vessel region obtained in 4), the length of the axial connected region of each blood vessel signal in the small blood vessel region image obtained in 2) and the large blood vessel region image obtained in 3) after removing background noise is restricted respectively. The size of each connected region is restricted to the corresponding transverse thickness after processing, so as to obtain the small blood vessel region image and the large blood vessel region image after removing axial artifacts. 6) Combine the small vessel region image obtained in 5) with the large vessel region image after preliminary removal of axial artifacts into a single 3D OCTA image after preliminary removal of axial artifacts; 7) The 3D OCTA image obtained in 6) with preliminary removal of axial artifacts is further enhanced with Hessian-filtering to obtain the final 3D OCTA image with axial artifacts removed.

2. The 3D OCTA axial artifact suppression method based on morphological features according to claim 1, characterized in that, In step 3), the specific threshold selection method is as follows: calculate the ratio of the average intensity of multiple background noise regions in the large blood vessel region to the average intensity of the small blood vessel region, and determine the maximum value among all ratios; select the product of a percentage slightly greater than the maximum value and the average intensity of the small blood vessel region as the intensity threshold of the large blood vessel region.

3. The 3D OCTA axial artifact suppression method based on morphological features according to claim 1, characterized in that, In step 4), the transverse thickness is processed as follows: for vessels that are not fully axial, the processed transverse thickness value TM is the transverse thickness T calculated by the thickness algorithm divided by the polar angle. The sine value and rounded down:

4. The 3D OCTA axial artifact suppression method based on morphological features according to claim 1, characterized in that, In step 5), the method for limiting the length of the axial connected domain of the blood vessel is as follows: taking the maximum intensity value of the axial connected domain of the blood vessel as the center, and limiting the number of voxels in the connected domains above and below the center to half of the transverse thickness value of the blood vessel after processing, TM / 2.

Images