Open Access
Add to BibSonomy
Abstract
Optical coherence tomography angiography (OCTA) images suffer from inevitable micromotion (breathing, heartbeat, and blinking) noise. These image artifacts can severely disturb the visibility of results and reduce accuracy of vessel morphological and functional metrics quantization. Herein, we propose a multiple wavelet-FFT algorithm (MW-FFTA) comprising multiple integrated processes combined with wavelet-FFT and minimum reconstruction that can be used to effectively attenuate motion artifacts and significantly improve the precision of quantitative information. We verified the fidelity of image information and reliability of MW-FFTA by the image quality evaluation. The efficiency and robustness of MW-FFTA was validated by the vessel parameters on multi-scene in vivo OCTA imaging. Compared with previous algorithms, our method provides better visual and quantitative results. Therefore, the MW-FFTA possesses the potential capacity to improve the diagnosis of clinical diseases with OCTA.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Optical coherence tomography (OCT) is a powerful non-invasive medical imaging technology that can be used to effectively acquire high-resolution three-dimensional (3D) tomographic images of tissues, and it has been extensively applied in clinical ophthalmic disease diagnosis and other medical studies [1,2]. Recently, with the development of laser and data acquisition technologies, a new imaging technique, OCT angiography (OCTA), which provides the advantages of rapid and volumetric imaging, has become popular in the field of retinal vascular diseases [3–5]. Based on the differences between static and dynamic states caused by the intensity and phase of the backscattered light emerging from the tissues, OCTA can be used to obtain microcirculation information in vivo without requiring an exogenous contrast agent such as fluorescent dyes [6]. OCTA, following its proposal, has quickly obtained clinical approval from the FDA in 2016 [7]. Our study is based on the typical OCT technology, which generates images by interferometrically measuring the amplitude and delay of reflected or backscattered light based on a modification of classic Michelson interferometry.
Complete OCTA mapping is constructed by computing decorrelation between sequential OCT B-scans, which are acquired by repeatedly scanning the same location [8–10]. The principle of OCTA is to distinguish the difference between two OCT signals, one is backscattered from static structural tissue and the other is backscattered from moving RBCs (Red Blood Cell) in vessels. As a phase-signal-based OCTA, the difference of OCT signal phase information in subsequent scans caused by the moving particles generates the angiographic contrast, providing the opportunity to visualize the microvasculature.
However, in practical application, owing to regular respiration and the involuntary movement of the measured object, the target imaging region is disturbed by micromotion. In addition, owing to the imaging speed limitation, these micromotions might change the distribution of signal intensities, consequently leading to high decorrelation similar to that observed in the microvasculature, which will generate regular stripe motion artifacts [11]. For endoscopic or in vivo imaging in particular, the motion stripe artifacts may significantly degrade the image quality and hinder quantitative analysis.
Previous studies have designed a few algorithms and hardware systems to attenuate the OCTA noise signal intensities [12]. These algorithms, including the block-matching and 4D collaborative filter (BM4D) and wavelet-based singular value decomposition (K-SVD) filter, which are mainly designed for speckle noise in OCTA images, negligibly affect stripe motion artifacts [13,14]. Frequency domain filtering using Fast Fourier Transform (FFT) [14] has been proposed to reduce stripe motion artifacts. However, it is challenging to eliminate stripe motion artifacts and protect original features simultaneously, because the human capacity to visually discriminate undesired textures significantly surpasses filter selectivity by far [13]. The motion-tracking system composed of the tracking LSO (Line Scan Ophthalmoscope) is able to guide the OCT scanning, which can also effectively remove the motion artifacts [15]. Generally, the software optimization is a relatively low-cost and convenient way for motion artifact removal.
Recently, wavelet-FFT, combining the condensation of the stripe information from wavelet and advantages of selective attenuation of the noise from frequency domain filtering, can theoretically solve the aforementioned problem [13]. However, for OCTA in vivo imaging, micromotion can cause large relative displacements between local imaging areas, which leads to stronger and wider stripe motion artifacts. Thus, it is challenging to achieve deep filtering via single wavelet-FFT, and a certain intensity of motion stripe artifacts remain. Conversely, the “blurring” comprising the moderately high signals compared with the background appears around the stripes after the wavelet-FFT, which is probably recognized as a vascular structure [16].
In this study, we proposed a new filtering method, multiple wavelet-FFT algorithm (MW-FFTA), comprising multiple integrated process combined with wavelet-FFT and minimum reconstruction. First, the minimum value between original and post-filtering images was calculated to effectively remove the “blur” and restore the original background [16,17]. Then, to achieve the optimized attenuation of stripe motion artifacts, we performed multiple integrated processes to simultaneously maintain the relative intensity between residual noise and background signal. Furthermore, to demonstrate the effectiveness of MW-FFTA for vascular networks in multi-scene imaging on a wide field of view (FOV), we qualitatively and quantitatively evaluated different objects (retina, ear, bladder, intestine, and tumor) before and after processing. Our results demonstrate that MW-FFTA can considerably enhance the OCTA imaging performance and microvasculature information extraction precision. We believe that the MW-FFTA is reliable and promising for biomedical and clinical applications of OCTA imaging.
2. Method
2.1 OCTA system
This research is based on a home-built swept-source OCT system. The schematic of the system is illustrated in Fig. 1. The system comprised a swept-source laser (Axsun 105, AXSUN Technologies Inc., Billerica, MA) with a central wavelength of 1060 nm, which operates at a 200 kHz sweep rate. The OCT signals from the balanced detector were sampled with a high-speed digitizer (400 MHz, 12-bit up to 1.8 GS/s, AlazarTech, ATS9360) using the built-in trigger and clock signals from the Axsun laser. This system had an axial resolution of approximately 5 µm in air and a lateral resolution of 29 µm. There were 1400 useful sample points for each interferogram. Owing to unintended reflections from fiber tips, sample/reference arm optics, and an unstable A-line trigger, a few fixed-pattern noise locations in the OCT B-scan images were present. One prominent fixed-pattern noise was around the zero-depth position. More details of system have been described in our previous work [18,19].
2.2 Theory and methods of wavelet-FFT
The stripe motion artifacts which are common in OCTA images present a significantly higher spatial frequency bandwidth along its width extension direction instead of a lower frequency bandwidth along its length extension direction. The frequency domain filter has generally been used to attenuate these stripe motion artifacts. The original image $f(x,y)$ in the time domain is transformed into $F(x,y)$ in the frequency domain via Fast Fourier Transform (FFT). Low-frequency components are transferred to the center through the centralization of the spectrum. According to the direction of stripe motion artifacts, a bandpass filter based on the Gaussian damping function is used to attenuate the components on horizontal or vertical axis [14]. However, detail characteristics of the original image are attenuated along stripes direction, which can lead to the loss of original vascular information. To overcome this limitation, it is required to combine wavelet transform to precisely extract and classify the detailed information of OCTA images
As shown in Fig. 2(a), the row signals of original images are filtered using high-pass and low-pass filters; then, the filtering results are subsampled into two components. Further, the column signals of these components are filtered using the same filters and subsampled again to acquire approximation, diagonal, vertical, and horizontal bands. Vertical bands possess low frequencies along columns and high frequencies along rows, so the low-pass filter along the columns and high-pass filter along the rows can be used to obtain vertical detail information [13]. The other bands are acquired based on the same principle. Three detail bands reflect the detail information along different directions, and the OCTA images can be reconstructed based on the four components in Fig. 2(b).
Fig. 2. Schematic diagram of wavelet decomposition to original OCTA images. (a) The flow diagram of the single-level wavelet decomposition. The 1$\downarrow $2 refers to a down-sampling of the rows while 2$\downarrow $1 refers to a down-sampling of the columns (b) Multiple-level wavelet decomposition accomplishes depth-sampling for multidirectional detail information.
As shown in Fig. 2(b), the approximation and detail information located in the lower frequency band can be obtained using multiple wavelet decomposition. Thus, the original OCTA images $f(x,y)$ can be decomposed using wavelet into the following four components:
${c_l}_{L,m,n}$ is the approximation band, ${c_h}_{l,m,n}$ is the horizontal band, ${c_v}_{l,m,n}$ is the vertical band and ${c_d}_{l,m,n}$ is the diagonal band. L is the number of iterations, $l \in \{{1, \ldots ,L} \}$.The vertical details were then transformed into the frequency domain for the further compression of vertical lines. A Gaussian function-based bandpass filter was built for stripe removal. Therefore, the wavelet-FFT filter is a combined method via applying FFT-based frequency domain filtering to extract the specific detailed component from the multiple wavelet decomposition of original images, which can attenuate stripe motion artifacts.
2.3 Theory and methods of MW-FFTA
Although the attenuation of stripe motion artifacts could be preliminarily achieved through wavelet-FFT as aforementioned, it is not a satisfactory filtering method for OCTA. According to the schematic diagram in Fig. 3(a), the motion artifact is attenuated by preliminary wavelet-FFT. However, a kind of high-signal noise named “blur” appears in the vicinity of the stripe motion artifacts, because the frequency domain filter could enhance the signal values adjacent to stripe motion artifacts [14,16,17]. We compare the preliminary wavelet-FFT filtering result with original image to take minimum signal value to remove the “blur” generated by filtering. Obviously, after removing the “blur”, motion artifact through preliminary attenuation is presented as relatively high signal value in the local area. Therefore, we achieve depth attenuation of motion artifacts and preserve the original image information by multiple wavelet-FFT and minimum reconstruction. Meanwhile, the effectiveness of each procedure in our method is demonstrated in Fig. 3(a) by the OCTA image with real motion artifacts. The red arrow marks the “blur” as described previously, which appears around the artifacts after wavelet-FFT and is removed by minimum reconstruction. The green arrow marks a representative motion artifact, which is deeply attenuated by multiple wavelet-FFT and minimum reconstruction.
Fig. 3. The schematic diagram and effect of MW-FFTA. (a) The demonstration of “blur” removing and artifacts attenuation with MW-FFTA. The red and green arrows point to representative “blur” and artifacts respectively. (b) Quantitative analysis of cross-section signal intensity on the yellow line. The 1-9 black arrows indicate the motion artifact information. The purple arrow indicates the appearance and removal of the representative “blur”. The blue arrows represent the microvascular information. (c) The model architecture and specifications of MW-FFTA.
As shown in Fig. 3(b), the 1-9 black arrows indicate the signal intensity of motion artifacts at the yellow cut-off line. The purple arrow indicates the appearance and removal of the representative “blur” between 4 and 5 motion artifacts. The blue arrow between 1 and 2 motion artifacts represents the smaller microvascular information and the blue arrow between 6 and 7 motion artifacts represents the stronger microvascular information. We could confirm that the signal intensity of “blur” is similar to that of the smaller microvasculature using ImageJ (2020 version), which potentially leads to deviations in the extraction of vessel information. Therefore, minimum reconstruction is adopted to effectively restore the original background. Simultaneously, minimum reconstruction creates relative differences between the original background and stripe motion artifacts, laying a foundation for the further attenuation of the residual artifacts.
Therefore, we propose a new filtering method that iterates wavelet-FFT and minimum reconstruction repeatedly to attenuate the stripe motion artifacts deeply and protect the vessel information. Figure 3(c) shows the complete algorithm flow of MW-FFTA, which has the advantage to deeply attenuate stripe motion artifacts in OCTA in vivo imaging while preserving the maximum possible vascular information.
In this study, all OCTA images are processed in MATLAB (R2016b, MathWorks, Inc.) on a personal computer (AMD Ryzen 7 4800U@1.80 GHz). The entire MW-FFTA process required less than 1.5 seconds. All OCTA images were acquired using a host computer (AMD Ryzen 3.6 GHz, 1 T RAM, X64 system) equipped with an 8-lane PCI Express Gen2 interface.
3. Result and discussion
3.1 Evaluation of filter performance with different parameters
To evaluate the MW-FFTA performance, as a hardware solution, the stabilizing device is used to acquire a high-quality OCTA image (400*400 pixels) of the ear without any noticeable stripe motion artifacts, as shown in Fig. 4(a). Therefore, the discrepancy between the simulated artifacts and filtered images can be observed by comparing with original images. Moreover, the performance differences in various filters are reflected through these comparisons. Further, we evaluated the MW-FFTA performance according to the image quality and precision of the vascular information extraction. As shown in Fig. 4(a), vertical noise stripes are modeled as white vertical lines. The width of the artificial vertical lines are randomized between 1 and 3 pixels, the length is randomized between 0.7 and 0.9 of the image length, and the pixel value of the lines are randomized between 150 and 180 [17].
Fig. 4. The demonstration of stimulated stripe motion artifacts and effect of MW-FFTA with different parameters. (a) The high-quality OCTA image acquired with stabilizing device, the image with simulated stripe motion artifacts, and the filtering results using MW-FFTA with different parameters. (b) The variation of Peak Signal to Noise Ratio (PSNR), Structural Similarity (SSIM) and Gradient Magnitude Similarity Deviation (GMSD) with changes of different filter parameters.
To further verify the influence of the wavelet decomposition grade $L$, Gaussian attenuation intensity $\sigma $, and the iteration time $N$ of MW-FFTA, we evaluated the results of filters with different parameters based on the simulated stripes shown in Fig. 4(a). The results of filtering indicate that L is closely related to width of the artifacts, and $\sigma $ is related to the attenuation intensity. Simultaneously, a certain range of $N$ can achieve different attenuation levels for residual stripe artifacts and realize a better filtering effect compared with that of wavelet-FFT.
3.2 Image quality evaluation
Subsequently, to investigate the potential application of medical images, the filter performance was evaluated via both human and computer vision. Further, three of the most widely used evaluation image quality metrics were applied to quantify the image distortion perception of humans. A superior image quality enables doctors to diagnose accurately. For computer vision, we extracted the parameters of blood vessels using computer vision and evaluate them by observing changes in these parameters.
Peak Signal to Noise Ratio (PSNR) indicates the relationship between image maximum signal and background noise, and it is calculated as follows:
Structural Similarity (SSIM) is another extensively used full-reference image quality evaluation index combining luminance, contrast, and structure comparison of images, and they are calculated as follows:
where $l$, $c$, and s are three independent attributes of the image. $\alpha $, $\beta $, and $\gamma $ are used to adjust the relative weights of the three attributes. SSIM, based on the assumption that human visual system is highly adaptable to extracting structural information from the viewing field, indicates the structural differences between the images. Therefore, it is a reliable reference to describe the perception of humans to the differences of images [21].Gradient Magnitude Similarity Deviation (GMSD) is a full-reference evaluation for images that only calculates the gradient amplitude of the images as a feature to generate high-precision quality prediction scores [22]. It is calculated as:
where ${m_a}$ and ${m_b}$ are the gradient magnitude of the original and the processed images, respectively. GMSM is the mean value of the GMS. OCTA images contain abundant local structural information of blood vessels, which degrades gradient amplitude during filtering. Therefore, we evaluated the local image quality by calculating the similarity of the local gradient amplitude, and measured the global quality change by calculating the standard deviation of the local image quality.Next, we used these three metrics to evaluate the differences in MW-FFTA with various parameters and the discrepancy of other filters. As shown in Fig. 4(b), PSNR, SSIM, and GMSD exhibit similar results as L changes. When $L \le 4$, the image quality significantly improves, and slightly degrades afterward. Moreover, the change in $\sigma $ can lead to the vibration of these three parameters, and $\sigma = 3$ or $\sigma = 5$ are appropriate values for filtering. The number of iterations, N, confirms that MW-FFTA processing acquires a superior image quality compared with single wavelet-FFT. Additionally, when $N = 3$, the results are optimal. However, the results also demonstrate that the filtering quality degrades with increasing N, indicating that extremely high iteration numbers hinder image quality.
Finally, as shown in Table 1, to evaluate the image quality, we quantitatively calculated the values of three metrics and compared the former filters (FFT, K-SVD and wavelet-FFT) with MW-FFTA. As listed in Table 1, all these filtering methods are effective for increasing the values of PSNR and SSIM and decreasing the value of GMSD. However, our method achieves the best results. That is, our quantitative results demonstrate the superiority of MW-FFTA from human's perception.
Table 1. Results of image quality evaluation in different filtering method (Means ± Std)
3.3 Evaluation of blood vessel information extraction
OCTA can be used to visualize the vessel information of the tissue, providing the advantage of predicting the development of retinal vascular disease [23], tumors [19], skin diseases, and other diseases in clinics [24,25]. However, the OCTA image information of microvasculature structure is significantly degraded by artifacts from motion (horizontal and vertical lines). Moreover, these artifacts could be mistaken for blood vessels. Therefore, they are considerable hindrances for the accurate information extraction of vessels, which could lead to the deviation of monitoring and diagnosis. Therefore, the variation of blood vessel information is another important indicator to evaluate the filter performance.
Herein, we selected the same OCTA image (400*400 pixels) of the ear with simulated stripe motion artifacts. To comprehensively evaluate the vessel information, the representative morphological parameters, Vessel Area Density (VAD), Vessel Skeleton Density (VSD), Vessel Diameter Index (VDI), Fractal Dimension (FD), and Lacunarity were selected to quantify blood vessel information [25–31].
As shown in Fig. 5(b), the original image is converted to a binary image using MATLAB (R2016b, MathWorks, Inc.) based on a hessian filter and an adaptive threshold [27,32,33]. This is an effective approach to achieve the preliminary segmentation of microvasculature, and important vessel information is illustrated in the binary image [26]. Then a vessel skeleton map is established, as shown in Fig. 5(c), where all the vessels are represented as pixel lines, regardless of their diameters. The length information of microvasculature can be extracted via skeletonization. Furthermore, to illustrate the complete vessel perimeter information, a vessel perimeter map was constructed via edge extraction, as shown in Fig. 5(d). Finally, the abovementioned five vessel information parameters were calculated based on these vessel maps.
Fig. 5. The quantitative analysis process of vasculature in OCTA image. (a) The original OCTA image. (b) The binary vessel image acquired using hessian filter and threshold segmentation. (c) The vessel vasculature skeleton map. (d) The vessel perimeter map based on edge extraction.
To evaluate the function of microvasculature, VAD was calculated as the ratio of vessel area to total area in the binary vessel map. Therefore, the VAD vibration is coupled with the changes in vessel length and size. VAD is an evaluation index for decreases in microvascular perfusion [23], and the equation is described as follows:
VSD is a method used for evaluating the vessel length without considering the vessel diameter information. It calculates the ratio of the area occupied by vasculature skeleton to the total area of the skeleton map [26]. Hence, it has been reported to be valuable in the analysis of retinal diseases such as diabetic retinopathy (DR) and age-related macular degeneration (AMD) [26]. VSD is expressed as follows:
VDI is a parameter of significance for measuring the average vessel diameter. Moreover, owing to its sensitivity to vasodilation, VDI can be used to effectively identify the localized vasodilation and provide a diagnostic reference of abnormal vessels. Once the stripe motion artifact is identified as a vascular structure, VDI deviate in length and diameter. It can be used to calculate the ratio of vessel area map to vessel skeleton map as follows:
Additionally, FD has wide-ranging applications in image compression and segmentation. Thus, during the quantitative analysis of OCTA vascular information, FD is a significant parameter used to indicate the complexity of blood vessels, which is closely related to the aggressiveness of the tumor region during OCTA clinical diagnosis. The accurate determination of FD can facilitate early and precise prediction of tumors [34]. FD is calculated as follows:
Lacunarity is another texture complexity evaluation parameter based on FD. While analyzing vasculature information, lacunarity reflects vessel heterogeneity. The higher the lacunarity, the greater the inhomogeneity the vessel area. It is expressed as follows:
where ${\sigma _{\varepsilon ,g}}$ and ${\mu _{\varepsilon ,g}}$ are the mean and standard deviation, respectively, from the entire pixels related to the scaled box size, $\varepsilon $, in each grid, $g$. Herein, we selected the Angiotool software (Angiotool64 0.6a version) to calculate lacunarity [36]. To summarize, we integrated the aforementioned parameters as an evaluation system to verify the filter performance by comparing various filtering methods.First, Fig. 6(a) shows the OCTA images processed using different filters as a visual representation of the filtering performance. These results indicated that it is more challenging to eliminate artifacts using the wavelet-FFT than other filters. In addition, the background is filled with “blur” in the results of FFT and wavelet-FFT. Figure 6(b-d) show the filtering differences during vascular information extraction. The results indicate that the incomplete removal of artifacts will be identified as vascular diameter and skeleton information, as well as truncate the original vessels, consequently resulting in the partial loss of vascular information. Moreover, the “blur” produced by the FFT and wavelet-FFT might be defined as microvascular to disturb extraction precision. Finally, as shown in Fig. 6(e), based on the VAD calculating method, we used the sliding kernel (15*15 pixels) to generate the down-sampling map, and it was smoothed using a 5 × 5 Gaussian filter to construct the vessel density map, which is a comprehensive representation of vessel aggregation rate.
Fig. 6. The performance evaluation process and effect demonstration of various filtering methods for stripe motion artifacts. (a) The original OCTA images. (b) The binary vessel images. (c) The vessel skeleton maps. (d) The vessel perimeter maps. (e) The vessel density maps acquired by a convolution kernel moving across binary vessel images.
The quantitative analysis results are listed in Table 2. As expected, a considerable difference is observed between the original and simulated stripes images in VAD, VSD, FD, and Lacunarity, because artifacts are inaccurately detected as vessels. Moreover, FD and Lacunarity, closely related to vascular complexity, decrease owing to artifacts, and increase owing to the “blur”. Notably, MW-FFTA achieves excellent results compared with those of other methods.
Table 2. Results of quantitative vessel analysis by using different filtering method (Means ± Std)
To summarize, compared with traditional filters, the MW-FFTA exhibited more accurate calculation results of VAD, VSD, VDI, FD, and lacunarity. Therefore, our developed MW-FFTA is of great significance to reduce the influence of stripe motion artifacts on the vascular information extraction in OCTA in vivo imaging, thereby improving diagnosis accuracy and diseases treatment.
3.4 MW-FFTA application in OCTA multi-scene in vivo imaging
To further verify the effectiveness of MW-FFTA in multi-scene, we selected the retina of human and various positions of live animal models for OCTA in vivo imaging shown in the Fig. 7(a). First, the retina is among the most widely used OCTA field, and the artifacts significantly limit the diagnostic accuracy for retinal diseases. As a highly transparent area, the mice ear is commonly used to construct the skin diseases and subcutaneous tumor models. In addition, organs such as the bladder and intestine are gradually becoming important observation specimens with the development of OCT endoscope. They have a stronger relative motion to the lens, resulting in significant motion artifacts, which existed in the results, thereby limiting OCTA application expansion. Furthermore, a new field of OCTA in vivo application, tumor growth monitoring and treatment combined with nano-contrast agents, is considerably adversely affected by the artifacts in the quantitative analysis of tumor microvasculature [19].
Fig. 7. The MW-FFTA application results in OCTA multi-scene in vivo imaging. (a) The demonstration of the experimental method. (b) The result demonstration of MW-FFTA using Frangi2D vascular structure reconstruction image, vascular skeleton image, and 3D model image of ROI (300*300 pixels) selected from OCTA original image (800*800 pixels) with stripe motion artifacts. The ROI are shown as green rectangles and tumor is shown as a white region. (c) The quantitative analysis of various vascular parameters under different filters.
We show the filtering results of multi-scene in OCTA application by MW-FFTA to demonstrate the superior performance of our method. As shown in Fig. 7(b), the original OCTA images comprise 800*800 pixels (10*10 mm). Then, we selected a region of interest (ROI) comprising 300*300 pixels, and constructed Frangi2D, vascular skeleton, and 3D models to analyze the vascular morphology to demonstrate the filtering effect.
In the original OCTA images of the five positions, the existing stripes are clear, and are even more distinct in the extracted ROI images. However, the versatility of MW-FFTA was fully confirmed, and the significant attenuation or even the disappearance of artifacts in various morphological analysis models is shown in Fig. 7(b). More importantly, high-quality morphology image can more accurately qualitatively analyze of microcirculation information.
Subsequently, to investigate the MW-FFTA accuracy and robustness, as shown in Fig. 7(c), we calculated the various vessel information of three ROI images in the five positions before and after different filtering methods. As shown in Fig. 7(c), the effectiveness of MW-FFTA is superior to those of other filter methods. As shown in Table 3, the variation of these parameters before and after filtering reflects the precision of vascular information extraction. In addition, we can observe that VAD, VSD, FD, and Lacunarity all obtained a higher precision under our MW-FFTA, and the improvement in VAD, VSD, and FD is distinct, and the increase can reach 20.44%. Notably, for Lacunarity, the improvement is more discernable, which could reach 76.23%. As for VDI, the advantages are relatively less compared with those of VAD, VSD, FD, and Lacunarity. It is potentially challenging to determine whether the variation is favorable or not because of the fewer reference images. Meanwhile, if the part of the vascular signal happens to be located on coverage area of motion artifacts, the intensity of these signals become lower with the attenuation of motion artifact by MW-FFTA. Therefore, in the vascular information extraction, choosing a suitable threshold segmentation may be helpful to restore the discontinuous vessels. Nevertheless, our results quantitatively demonstrate the significance of MW-FFTA for OCTA in vivo imaging.
Table 3. Improvements in the precision of vascular information extraction under various methods (Means ± Std)
We selected female BALB/c mice purchased from Shanghai Slac Laboratory Animal Co., Ltd as experimental animals. Experimental protocols involving animals were approved by the Animal Care and Use Committee of Xiamen University and were licensed by the Department of Science and Technology of Fujian, the People’s Republic of China. (SPHIRB-202102). During the OCTA imaging, the mice were fixed on a high precision translation stage to acquire the optimal imaging location. A heating pad was utilized to maintain the body temperature of the mouse during imaging. For retina, ear and subcutaneous tumors, we placed targets in the effective area of OCTA to acquire in vivo imaging data. For the intestine and bladder, their locations in the depth direction exceed the imaging depth of OCTA. Therefore, we created imaging windows to facilitate OCTA data acquisition.
4. Conclusion
To summarize, we developed a novel MW-FFTA comprising multiple integrated process combined with wavelet-FFT and minimum reconstruction. It was significantly effective to suppress motion stripe artifacts widely existing in OCTA in vivo imaging. Moreover, we demonstrated that the image quality and accuracy of OCTA vessel information extraction can be considerably improved using our method. Furthermore, we explored the effect of filter parameters on its performance to provide valuable insights into filtering. Notably, the versatility of MW-FFTA was demonstrated in multi-scene in vivo OCTA imaging. More importantly, we evaluated the performance of different filters through various vessel parameters, and the results demonstrated that MW-FFTA was superior for OCTA in vivo imaging from both human and computer vision. These results encourage us to potentially apply this method clinically for OCTA ophthalmology and other biomedical imaging.
Funding
Basic and Applied Basic Research Foundation of Guangdong Province (2021A1515011654); Fundamental Research Funds for the Central Universities (20720210117); National Natural Science Foundation of China (61675134, 62175156); Key Laboratory of OptoElectronic Science and Technology for Medicine of Ministry of Education, Fujian Provincial Key Laboratory of Photonics Technology (JYG2105); Xiamen Key Laboratory of Endocrine-Related Cancer Precision Medicine (XKLEC2021KF03); XMU Undergraduate Innovation and Entrepreneurship Training Programs (2021X1119, 2021Y1119, 202210384051, S202110384391, S202210384404); Shenzhen Bay Laboratory (SZBL2019062801005).
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. A. H. Kashani, C. L. Chen, J. K. Gahm, F. Zheng, G. M. Richter, P. J. Rosenfeld, Y. Shi, and R. K. Wang, “Optical coherence tomography angiography: A comprehensive review of current methods and clinical applications,” Prog. Retinal Eye Res. 60, 66–100 (2017). [CrossRef]
2. D. Huang, E. A. Swanson, C. P. Lin, J. S. Schuman, W. G. Stinson, W. Chang, M. R. Hee, T. Flotte, K. Gregory, C. A. Puliafito, and J. G. Fujimoto, “Optical Coherence Tomography,” Science 254(5035), 1178–1181 (1991). [CrossRef]
3. T. T. Hormel, Y. Jia, Y. Jian, T. S. Hwang, S. T. Bailey, M. E. Pennesi, D. J. Wilson, J. C. Morrison, and D. Huang, “Plexus-specific retinal vascular anatomy and pathologies as seen by projection-resolved optical coherence tomographic angiography,” Prog. Retinal Eye Res. 80, 100878 (2021). [CrossRef]
4. H. Khalid, R. Schwartz, L. Nicholson, J. Huemer, M. H. El-Bradey, D. A. Sim, P. J. Patel, K. Balaskas, R. D. Hamilton, P. A. Keane, and R. Rajendram, “Widefield optical coherence tomography angiography for early detection and objective evaluation of proliferative diabetic retinopathy,” Br. J. Ophthalmol. 105(1), 118–123 (2021). [CrossRef]
5. H. L. Takusagawa, L. Liu, K. N. Ma, Y. Jia, S. S. Gao, M. Zhang, B. Edmunds, M. Parikh, S. Tehrani, J. C. Morrison, and D. Huang, “Projection-Resolved Optical Coherence Tomography Angiography of Macular Retinal Circulation in Glaucoma,” Ophthalmology 124(11), 1589–1599 (2017). [CrossRef]
6. Y. Cui, Y. Zhu, J. C. Wang, Y. Lu, R. Zeng, R. Katz, F. Vingopoulos, R. Le, I. Lains, D. M. Wu, D. Eliott, D. G. Vavvas, D. Husain, J. W. Miller, L. A. Kim, and J. B. Miller, “Comparison of widefield swept-source optical coherence tomography angiography with ultra-widefield colour fundus photography and fluorescein angiography for detection of lesions in diabetic retinopathy,” Br. J. Ophthalmol. 105(4), 577–581 (2021). [CrossRef]
7. R. F. Spaide, J. G. Fujimoto, N. K. Waheed, S. R. Sadda, and G. Staurenghi, “Optical coherence tomography angiography,” Prog. Retinal Eye Res. 64, 1–55 (2018). [CrossRef]
8. C. L. Chen and R. K. Wang, “Optical coherence tomography based angiography [Invited],” Biomed. Opt. Express 8(2), 1056–1082 (2017). [CrossRef]
9. J. K. Barton and S. Stromski, “Flow measurement without phase information in optical coherence tomography images,” Opt. Express 13(14), 5234–5239 (2005). [CrossRef]
10. A. Mariampillai, B. A. Standish, E. H. Moriyama, M. Khurana, N. R. Munce, M. K. K. Leung, J. Jiang, A. Cable, B. C. Wilson, I. A. Vitkin, and V. X. D. Yang, “Speckle variance detection of microvasculature using swept-source optical coherence tomography,” Appl. Phys. Lett. 33(13), 1530–1532 (2008). [CrossRef]
11. A. Li, G. Zeng, C. Du, H. Zhang, and Y. Pan, “Automated motion-artifact correction in an OCTA image using tensor voting approach,” Appl. Phys. Lett. 113(10), 101102 (2018). [CrossRef]
12. J. Yang, Y. Hu, L. Fang, J. Cheng, and J. Liu, “Universal digital filtering for denoising volumetric retinal OCT and OCT angiography in 3D shearlet domain,” Opt. Lett. 45(3), 694–697 (2020). [CrossRef]
13. B. Münch, P. Trtik, F. Marone, and M. Stampanoni, “Stripe and ring artifact removal with combined wavelet — Fourier filtering,” Opt. Express 17(10), 8567–8591 (2009). [CrossRef]
14. G. L. Guozhong Liu and R. W. Ruikang Wang, “Stripe motion artifact suppression in phase-resolved OCT blood flow images of the human eye based on the frequency rejection filter,” Chin. Opt. Lett. 11(3), 031701 (2013). [CrossRef]
15. Q. Zhang, Y. Huang, T. Zhang, S. Kubach, L. An, M. Laron, U. Sharma, and R. Wang, “Wide-field imaging of retinal vasculature using optical coherence tomography-based microangiography provided by motion tracking,” J. Biomed. Opt. 20(6), 066008 (2015). [CrossRef]
16. R. A. Byers, “Clinical Applications of Angiographic Optical Coherence Tomography,” (2018).
17. R. Byers and S. Matcher, “Attenuation of stripe artifacts in optical coherence tomography images through wavelet-FFT filtering,” Biomed. Opt. Express 10(8), 4179–4189 (2019). [CrossRef]
18. Y. Huang, M. Li, D. Huang, Q. Qiu, W. Lin, J. Liu, W. Yang, Y. Yao, G. Yan, N. Qu, V. V. Tuchin, S. Fan, G. Liu, Q. Zhao, and X. Chen, “Depth-Resolved Enhanced Spectral-Domain OCT Imaging of Live Mammalian Embryos Using Gold Nanoparticles as Contrast Agent,” Small 15(35), 1902346 (2019). [CrossRef]
19. D. Huang, G. Wang, J. Mao, C. Liu, Z. Fan, Y. Zhang, B. Zhang, Y. Zhao, C. Dai, Y. He, H. Ma, G. Liu, X. Chen, and Q. Zhao, “Intravital Whole-Process Monitoring Thermo-Chemotherapy Via 2D Silicon Nanoplatform: A Macro Guidance and Long-Term Microscopic Precise Imaging Strategy,” Adv. Sci. 8(16), 2101242 (2021). [CrossRef]
20. Q. Huynh-Thu and M. Ghanbari, “Scope of validity of PSNR in image/video quality assessment,” Electron. Lett. 44(13), 800–801 (2008). [CrossRef]
21. Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,” IEEE Trans. on Image Process. 13(4), 600–612 (2004). [CrossRef]
22. W. Xue, L. Zhang, X. Mou, and A. C. Bovik, “Gradient Magnitude Similarity Deviation: A Highly Efficient Perceptual Image Quality Index,” IEEE Trans. on Image Process. 23(2), 684–695 (2014). [CrossRef]
23. Y. Jia, S. T. Bailey, T. S. Hwang, S. M. McClintic, S. S. Gao, M. E. Pennesi, C. J. Flaxel, A. K. Lauer, D. J. Wilson, J. Hornegger, J. G. Fujimoto, and D. Huang, “Quantitative optical coherence tomography angiography of vascular abnormalities in the living human eye,” Proc. Natl. Acad. Sci. U. S. A. 112(8), 2395–2400 (2015). [CrossRef]
24. A. J. Deegan, F. Talebi-Liasi, S. Song, Y. Li, J. Xu, S. Men, M. M. Shinohara, M. E. Flowers, S. J. Lee, and R. K. Wang, “Optical coherence tomography angiography of normal skin and inflammatory dermatologic conditions,” Lasers Surg. Med. 50(3), 183–193 (2018). [CrossRef]
25. M. Casper, H. Schulz-Hildebrandt, M. Evers, R. Birngruber, D. Manstein, and G. Huttmann, “Optimization-based vessel segmentation pipeline for robust quantification of capillary networks in skin with optical coherence tomography angiography,” J. Biomed. Opt. 24(04), 1–11 (2019). [CrossRef]
26. Z. Chu, J. Lin, C. Gao, C. Xin, Q. Zhang, C. L. Chen, L. Roisman, G. Gregori, P. J. Rosenfeld, and R. K. Wang, “Quantitative assessment of the retinal microvasculature using optical coherence tomography angiography,” J. Biomed. Opt. 21(6), 066008 (2016). [CrossRef]
27. R. Reif, J. Qin, L. An, Z. Zhi, S. Dziennis, and R. Wang, “Quantifying optical microangiography images obtained from a spectral domain optical coherence tomography system,” Int. J. Biomed. Imaging 2012, 1–11 (2012). [CrossRef]
28. A. Y. Kim, Z. Chu, A. Shahidzadeh, R. K. Wang, C. A. Puliafito, and A. H. Kashani, “Quantifying Microvascular Density and Morphology in Diabetic Retinopathy Using Spectral-Domain Optical Coherence Tomography Angiography,” Invest. Ophthalmol. Visual Sci. 57(9), OCT362 (2016). [CrossRef]
29. S. Ghavami, M. Bayat, M. Fatemi, and A. Alizad, “Quantification of Morphological Features in Non-Contrast-Enhanced Ultrasound Microvasculature Imaging,” IEEE Access 8, 18925–18937 (2020). [CrossRef]
30. R. Ternifi, Y. Wang, E. C. Polley, R. T. Fazzio, M. Fatemi, and A. Alizad, “Quantitative Biomarkers for Cancer Detection Using Contrast-Free Ultrasound High-Definition Microvessel Imaging: Fractal Dimension, Murray's Deviation, Bifurcation Angle & Spatial Vascularity Pattern,” IEEE Trans. Med. Imaging 40(12), 3891–3900 (2021). [CrossRef]
31. M. Al-Sheikh, N. Phasukkijwatana, R. Dolz-Marco, M. Rahimi, N. A. Iafe, K. B. Freund, S. R. Sadda, and D. Sarraf, “Quantitative OCT Angiography of the Retinal Microvasculature and the Choriocapillaris in Myopic Eyes,” Invest. Ophthalmol. Visual Sci. 58(4), 2063–2069 (2017). [CrossRef]
32. A. Y. Kim, D. C. Rodger, A. Shahidzadeh, Z. Chu, N. Koulisis, B. Burkemper, X. Jiang, K. L. Pepple, R. K. Wang, C. A. Puliafito, N. A. Rao, and A. H. Kashani, “Quantifying Retinal Microvascular Changes in Uveitis Using Spectral-Domain Optical Coherence Tomography Angiography,” Am. J. Ophthalmol. 171, 101–112 (2016). [CrossRef]
33. A. F. Frangi, W. J. Niessen, K. L. Vincken, and M. A. Viergever, “Multiscale Vessel Enhancement Filtering,” IEEE Trans. Biomed. Eng. 1496, 130–137 (2000). [CrossRef]
34. O. S. Al-Kadi and D. Watson, “Texture analysis of aggressive and nonaggressive lung tumor CE CT images,” IEEE Trans. Biomed. Eng. 55(7), 1822–1830 (2008). [CrossRef]
35. C. A. Schneider, W. S. Rasband, and K. W. Eliceiri, “NIH Image to ImageJ: 25 years of image analysis,” Nat. Methods 9(7), 671–675 (2012). [CrossRef]
36. E. Zudaire, L. Gambardella, C. Kurcz, and S. Vermeren, “A computational tool for quantitative analysis of vascular networks,” PLoS One 6(11), e27385 (2011). [CrossRef]
