A method for improving the accuracy of blood vessel information extraction in OCTA live imaging

Abstract

The application provides a method for improving the precision of blood vessel information extraction in OCTA live imaging, comprising the following steps: performing two-dimensional wavelet decomposition on an image to obtain a low-pass frequency component, a horizontal structure detail component, a vertical structure detail component and a diagonal structure detail component; selecting the vertical structure detail component to perform fast Fourier transform to obtain a frequency spectrum diagram of the vertical structure detail component; performing frequency domain filtering on the frequency spectrum diagram of the vertical structure detail component by using a band-pass filter to obtain a filtered frequency domain signal; performing inverse fast Fourier transform on the filtered frequency domain signal and wavelet inverse transform to reconstruct a filtered image; comparing each pixel point in the reconstructed filtered image and the original image and taking the minimum value; iteratively performing the above process multiple times to reconstruct a deep denoising image; and the application can effectively remove the stripe motion artifacts existing in the OCTA live imaging result, and has very good reservation capability for image quality and blood vessel information.

Description

Technical Field

[0001] This invention relates to the field of OCTA in vivo imaging, and in particular to a method for improving the accuracy of vascular information extraction in OCTA in vivo imaging. Background Technology

[0002] Optical coherence tomography (OCTA) is a non-invasive angiography imaging technique developed in recent years based on OCT. Since its introduction, it quickly received FDA clinical approval in 2016 and has been widely used for the monitoring and diagnosis of diseases in the fundus and skin areas. Simultaneously, based on the development of OCT endoscopic technology, OCTA is also gradually being applied to the monitoring of living organs.

[0003] However, due to the unique OCTA image acquisition method, the final imaging result is a composite of multiple cross-sectional scans. Therefore, OCTA imaging inevitably involves a certain image acquisition time. Previous studies have shown that during image acquisition, the respiration of live animals causes micro-motions such as vibration, drift, or flickering. These micro-motions are the main source of image artifacts in OCTA live imaging results, specifically manifesting as stripe motion artifacts. These artifacts, with signal intensities similar to those of blood vessels, greatly interfere with subsequent vascular information extraction and quantitative analysis, severely impacting clinical medical diagnosis. Summary of the Invention

[0004] The main objective of this invention is to overcome the aforementioned deficiencies in the prior art and propose a method for improving the accuracy of vascular information extraction in OCTA in vivo imaging. This method significantly optimizes the vascular morphology and structure analysis and quantitative analysis of vascular information in OCTA in vivo imaging results from various sites, and is particularly effective against high-intensity, wide-sized stripe motion artifacts that often occur during in vivo organ imaging.

[0005] The present invention adopts the following technical solution:

[0006] A method for improving the accuracy of vascular information extraction in OCTA in vivo imaging includes the following steps:

[0007] S1: Perform two-dimensional wavelet decomposition on the image to obtain low-pass frequency components, horizontal structural detail components, vertical structural detail components, and diagonal structural detail components.

[0008] S2: Select the vertical structural detail components and perform a fast Fourier transform to obtain the spectrum of the vertical structural detail components;

[0009] S3: Use a bandpass filter to perform frequency domain filtering on the spectrum of the vertical structure detail components to obtain the filtered frequency domain signal;

[0010] S4: Reconstruct the filtered image by performing inverse fast Fourier transform and inverse wavelet transform on the filtered frequency domain signal;

[0011] S5: Compare each pixel in the reconstructed filtered image with the original image and take the minimum value;

[0012] S6: Iterate through S1-S5 multiple times to reconstruct the deep denoised image.

[0013] Specifically, the image is decomposed using two-dimensional wavelet decomposition to obtain low-pass frequency components, horizontal structural detail components, vertical structural detail components, and diagonal structural detail components, as follows:

[0014] The two-dimensional image signal f(x,y) is subjected to wavelet decomposition of the maximum order L, W={c lL,m,n ,c hl,m,n ,c vl,m,n ,c dl,m,n} and l∈{1,...,L}, resulting in four wavelet components, where c lL,m,n For low-pass frequency components, c hl,m,n For the horizontal structural detail component, c vl,m,n For the vertical structural detail component, c dl,m,n This represents the diagonal structural detail component.

[0015] Specifically, a bandpass filter is used to perform frequency domain filtering on the spectrum of the vertical structure detail components to obtain the filtered frequency domain signal; specifically:

[0016] The high-frequency signal at the center of the bandpass filter in the spectrum of the vertical structure detail component is moved to the edge of the spectrum, while the low-frequency signal is moved to the center of the spectrum; that is, the spectrum centering operation of moving the zero frequency point to the center of the spectrum is completed.

[0017] The signal on the horizontal axis of the spectrum after the spectrum centering operation is then attenuated by a filter based on the Gaussian damping function, where σ is the attenuation intensity.

[0018] Specifically, the filtered frequency domain signal is reconstructed by performing inverse fast Fourier transform and inverse wavelet transform, including:

[0019] The filtered frequency domain signal is first subjected to inverse spectrum centering to restore the zero frequency point position, and then subjected to inverse fast Fourier transform.

[0020] Combined with c lL,m,n For low-pass frequency components, c hl,m,n For horizontal structural detail components and c dl,m,n For the diagonal structural detail components, perform inverse wavelet transform W * ={c lL,m,n ,c hl,m,n ,c* vl,m,n ,c dl,m,n}, reconstruct the initially denoised image f(x,y) * .

[0021] Specifically, the depth-denoised image is reconstructed through multiple iterations S1-S5, as follows:

[0022] The process involves multiple iterations of wavelet-FFT and minimum value restoration of the image, with N iterations, to obtain the final filtered image f(x,y) after removing depth attenuation fringes and residual artifacts. * N .

[0023] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. Wavelet decomposition can downsample the original image through multiple channels to obtain image detail information from different orientations.

[0025] 2. A two-dimensional Fourier filter constructed from Fast Fourier Transform can rapidly, stably, and accurately attenuate the signal intensity of the fringe motion artifact component on the horizontal axis of the frequency domain.

[0026] 3. The combination of wavelet filters and Fourier filters can effectively separate stripe motion artifacts from blood vessels and background in the original image, can initially attenuate the noise signal intensity of stripe motion artifacts, and can preserve the original image information to a certain extent.

[0027] 4. By using the minimum pixel value repair method for both the reconstructed image and the original image, the microvascular structure and background blurred by the filter around the stripes can be repaired. At the same time, the high signal blur artifacts caused by the filter are removed while repairing the vascular structure.

[0028] 5. The combined process of N iterations of wavelet-FFT and minimum value repair can deeply attenuate motion artifact noise signals. This method has a significant optimization effect on OCTA live imaging results of various parts of the body for vascular morphology and structure analysis and quantitative analysis of vascular information. It is especially effective for high-intensity, wide-size stripe motion artifacts that often occur in live organ imaging. Attached Figure Description

[0029] Figure 1 A detailed flowchart of the noise reduction method provided in this embodiment of the invention;

[0030] Figure 2 A schematic diagram illustrating the wavelet decomposition used to extract image directional signal components in the denoising method provided in this embodiment of the invention;

[0031] Figure 3A schematic diagram of the frequency domain filtering applied in the denoising method provided in the embodiments of the present invention;

[0032] Figure 4 The denoising method provided in this embodiment of the invention is applied to experimental results of different OCTA in vivo imaging sites, and is represented by vascular morphology and structure.

[0033] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Detailed Implementation

[0034] Unlike speckle noise, which is common in medical image analysis, fringe motion artifact noise in OCTA live imaging has a clear directionality, making it difficult for filters targeting random noise to achieve good results. Wavelet Transform (WT), a novel transform analysis method, is characterized by its ability to fully highlight certain features of the signal through transformation. It can perform multi-scale refined sampling of the input image signal, automatically adapting to the requirements of time-frequency signal analysis, thus focusing on detailed information in various directions of the image. This overcomes some of the shortcomings of the Fourier Transform, representing a significant breakthrough in scientific methodology since the Fourier Transform.

[0035] For frequency domain filtering, the image spectrum can be obtained through Fast Fourier Transform (FFT). The edges of the original image are abrupt changes, and therefore represent high-frequency components in the spectrum. In most cases, noise in the original image is mostly high-frequency, while image information is mostly low-frequency. Frequency domain filtering transforms the image from a grayscale distribution to a frequency distribution to observe its characteristics and attenuate high-frequency noise components, achieving fast and accurate filtering.

[0036] Since OCTA fringe motion artifacts are mostly superimposed, they are always brighter than the underlying true signal. Therefore, a simple way to prevent the image signal around the fringe motion artifacts from being "blurred" by the high signal after filtering is to calculate the minimum value between the image before and after filtering. This step, which occurs after each wavelet-FFT filter, greatly restores the image information around the fringe.

[0037] Although the artifacts in the image after initial denoising using wavelet-FFT are significantly reduced, some residual artifacts still remain. Furthermore, due to the effect of minimum background inpainting, the background image information around the stripes is restored, and the residual stripe artifacts, compared to the background, once again become detail stripes with a certain contrast. Therefore, the combined process of multiple iterations of wavelet-FFT and minimum image inpainting deeply attenuates the residual stripe artifact noise signal, resulting in a better filtering effect.

[0038] Based on this, the present invention is proposed:

[0039] like Figure 1 The present invention provides a method for improving the accuracy of vascular information extraction in OCTA in vivo imaging, comprising the following steps:

[0040] S1: Perform two-dimensional wavelet decomposition on the image to obtain low-pass frequency components, horizontal structural detail components, vertical structural detail components, and diagonal structural detail components.

[0041] like Figure 2 The diagram shows the acquisition of the OCTA two-dimensional image signal f(x,y) and the wavelet decomposition W = {c} with a maximum order of L and a filter type of Daubechies. lL,m,n ,c hl,m,n ,c vl,m,n ,c dl,m,n Given that l∈{1,...,L}, we obtain four wavelet components representing the structural information of the grayscale image f(x,y), namely, the low-pass frequency band c lL,m,n Horizontal structural details c hl,m,n Vertical structural details c vl,m,n diagonal structural details c dl,m,n

[0042] S2: Select the vertical structural detail components and perform a fast Fourier transform to obtain the spectrum of the vertical structural detail components;

[0043] like Figure 3 As shown, for vertical structural details c vl,m,n Perform a Fast Fourier Transform (FFT) to obtain a spectrogram of the vertical structural details of the image.

[0044] S3: Use a bandpass filter to perform frequency domain filtering on the spectrum of the vertical structure detail components to obtain the filtered frequency domain signal;

[0045] like Figure 3 As shown, the zero-frequency point in the spectrum obtained in step 2) is moved to the middle of the spectrum, and the frequency domain signal of the stripe noise artifact is moved to a narrow band on the horizontal axis of the spectrum. Then, the narrow band signal on the horizontal axis of the obtained spectrum is attenuated to a certain extent by passing it through a filter based on the Gaussian damping function g(x,y), where the attenuation intensity is σ.

[0046] S4: Reconstruct the filtered image by performing inverse fast Fourier transform and inverse wavelet transform on the filtered frequency domain signal;

[0047] By performing an inverse spectrum centering operation on the attenuated spectrum in S3, the zero-frequency point position is restored. Then, the denoised vertical structure detail component c is obtained by performing an inverse fast Fourier transform (IFFT). * vl,m,nThen, through the inverse transform W of the wavelet type Daubechies * ={c lL,m,n ,c hl,m,n ,c * vl,m,n ,c dl,m,n}, thus obtaining the reconstructed image f(x,y) after preliminary removal of fringe motion artifacts. * .

[0048] S5: Compare each pixel in the reconstructed filtered image with the original image and take the minimum value;

[0049] Because the simple wavelet-FFT filter attenuates stripe motion artifacts, it causes high-signal blurring at the stripe edges relative to the background signal. This existing blurring can obscure parts of the microvascular structure and may also cause high-signal areas with similar intensity to blood vessels to appear in low-signal background regions where no blood vessel signals are present. This can significantly interfere with vascular morphology and quantitative analysis. Therefore, this method reconstructs the image f(x,y). * The method of comparing each pixel value with the pixel values ​​of the original image f(x,y) and taking the minimum value is used to make the image signal around the stripes consistent with the original image. Figure 1 This achieves the goal of repairing the blurred microvascular structure and background around the stripes.

[0050] S6: Iterate through S1-S5 multiple times to reconstruct the deep denoised image.

[0051] Although the stripes are significantly attenuated after initial denoising using simple wavelet-FFT, residual artifacts remain. Minimum background inpainting restores the background surrounding the stripes. Therefore, the residual stripe artifacts, compared to the restored background, reappear as detail stripes with some contrast. Through a combined process of N iterations of wavelet-FFT and minimum inpainting, the depth attenuation of these residual stripe artifacts is achieved. The final depth-filtered image f(x,y) is obtained. * N .

[0052] The above filtering method was validated by adding simulated noise to OCTA images without stripe motion artifacts. Simultaneously, suitable filter parameters were determined.

[0053] The effectiveness of the filtering is verified primarily through image quality assessment and the accuracy of vascular information extraction. Image quality assessment mainly uses Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) to evaluate whether the filter parameters cause image quality loss. The accuracy analysis of vascular information extraction focuses on evaluating two vascular information parameters: percentage of vessel area and total vessel length.

[0054] Table 1 shows two image quality assessment parameter values ​​for the filtering results of simulated fringe motion artifacts under the sole effect of filter parameter N. These two parameter values ​​represent a significant improvement compared to the original fringe pattern. Furthermore, these two assessment parameters can be used to determine the most suitable filter parameters for image quality under this fringe pattern.

[0055] Table 1

[0056]

[0057] Table 2 shows the vascular information extraction results of the simulated stripe motion artifact under different filter iteration numbers N. As can be seen from the table, for the two vascular parameters, the stripe motion artifact did not cause significant deviation; instead, wavelet-FFT filtering resulted in a noticeable deviation. However, both parameters gradually approached the original image as the filter iteration number N increased, reaching their closest approximation when N=7.

[0058] Table 2

[0059]

[0060] like Figure 4 As shown, OCTA images acquired from two live imaging sites—the fundus and the ear—were selected. The original images contained motion artifacts of varying widths in the stripes; therefore, we used the above-described embodiment to filter them and verify their noise reduction effect.

[0061] like Figure 4 As shown in the figure, the impact of the stripe motion artifact on the analysis of vascular morphology is first visually demonstrated, and the effect of this filter on stripe removal and preservation of vascular structure is also visually demonstrated.

[0062] Table 3 shows the results. Figure 4 The quantitative analysis results of the extraction of two types of vascular information parameters show significant differences between the various vascular information extracted from the original OCTA images of the two imaging organs and the extraction results from the images processed by the combined filter of this multi-iteration wavelet-FFT and minimum value restoration. This quantitatively demonstrates the significant effect of the filter.

[0063] Table 3

[0064]

[0065] As can be seen, the method employed in this invention comprehensively utilizes the advantages of wavelet filters in extracting detailed features in all directions of the image, the advantages of frequency domain filters based on fast Fourier transform in accurately attenuating high-frequency signals of image striated motion artifacts, the advantages of minimum pixel value repair of microvascular structures and background, and the advantages of the combined process of N-iteration wavelet-FFT and minimum value image repair in deeply attenuating residual signals of striated motion artifact filtering. This effectively removes striated motion artifacts in OCTA live imaging results while possessing excellent image quality and vascular information preservation capabilities. The effectiveness of this filtering method has been qualitatively and quantitatively verified from both vascular morphology and information extraction perspectives. Therefore, in the future, it can improve the imaging quality of OCTA in clinical ophthalmology, dermatology, and endoscopy, helping clinicians obtain more accurate vascular information data and promoting the clinical translation of OCTA.

[0066] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.

Claims

1. A method for improving the accuracy of extracting blood vessel information in OCTA live imaging, characterized in that, The method comprises the following steps: S1: performing two-dimensional wavelet decomposition on the image to obtain a low-pass frequency component, a horizontal structure detail component, a vertical structure detail component, and a diagonal structure detail component; S2: selecting the vertical structure detail component to perform fast Fourier transform to obtain a frequency spectrum diagram of the vertical structure detail component; S3: performing frequency domain filtering on the frequency spectrum diagram of the vertical structure detail component by using a band-pass filter to obtain a filtered frequency domain signal; S4: performing inverse fast Fourier transform on the filtered frequency domain signal and performing wavelet inverse transform to reconstruct a filtered image; S5: comparing each pixel point in the reconstructed filtered image and the original image and taking a minimum value; S6: iteratively performing S1-S5 to reconstruct a deep denoising image. The method comprises the following steps: The method comprises the following steps: The signal on the horizontal axis of the spectrum after the spectrum centering operation is further attenuated by a filter based on a Gaussian damping function, wherein is the attenuation intensity; The method comprises the following steps: The method comprises the following steps: combining for low-pass frequency components, for horizontal structure detail components, for diagonal structure detail components, and for denoised vertical structure detail components, inverse wavelet transform , reconstructing a preliminary denoised image .

2. The method for improving the accuracy of blood vessel information extraction in OCTA live imaging according to claim 1, characterized in that, The method comprises the following steps: Two-dimensional image signal a maximum number of times wavelet decomposition and four wavelet components are obtained, wherein is a low-pass frequency component, is a horizontal structure detail component, is a vertical structure detail component, is a diagonal structure detail component.

3. The method of claim 1, wherein, The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The The comprehensive process of multiple iterations of wavelet-FFT and minimum repair image, the number of iterations is , depth attenuation stripe residual artifacts, get the final filtered image .

Images