Open Access
Add to BibSonomy
Abstract
Port wine stain (PWS) is a congenital cutaneous capillary malformation composed of ecstatic vessels, while the microstructure of these vessels remains largely unknown. Optical coherence tomography angiography (OCTA) serves as a non-invasive, label-free and high-resolution tool to visualize the 3D tissue microvasculature. However, even as the 3D vessel images of PWS become readily accessible, quantitative analysis algorithms for their organization have mainly remained limited to analysis of 2D images. Especially, 3D orientations of vasculature in PWS have not yet been resolved at a voxel-wise basis. In this study, we employed the inverse signal-to-noise ratio (iSNR)-decorrelation (D) OCTA (ID-OCTA) to acquire 3D blood vessel images in vivo from PWS patients, and used the mean-subtraction method for de-shadowing to correct the tail artifacts. We developed algorithms which mapped blood vessels in spatial-angular hyperspace in a 3D context, and obtained orientation-derived metrics including directional variance and waviness for the characterization of vessel alignment and crimping level, respectively. Combining with thickness and local density measures, our method served as a multi-parametric analysis platform which covered a variety of morphological and organizational characteristics at a voxel-wise basis. We found that blood vessels were thicker, denser and less aligned in lesion skin in contrast to normal skin (symmetrical parts of skin lesions on the cheek), and complementary insights from these metrics led to a classification accuracy of ∼90% in identifying PWS. An improvement in sensitivity of 3D analysis was validated over 2D analysis. Our imaging and analysis system provides a clear picture of the microstructure of blood vessels within PWS tissues, which leads to a better understanding of this capillary malformation disease and facilitates improvements in diagnosis and treatment of PWS.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Port wine stain (PWS) is a common congenital skin vascular lesion, most of which occurs in the head and neck areas of patients, affecting the appearance and self-confidence of patients [1–3]. PWS is caused by vascular malformations that lead to flat and pink macules on the skin. The color of the macules will gradually become darker over time, and the macules will gradually become larger and thicker. At present, the common treatment method for PWS is photodynamic therapy (PDT) [4–6], although there are still cases where laser treatment fails, macules thickening or nodules and other symptoms requiring surgical treatments [7,8], and long-term postoperative follow-up observation of treatment response after surgery is also required. Proper diagnosis and treatment of PWS require visualizing these abnormal blood vessels within tissues. Therefore, there is an urgent need for non-invasive, high-resolution and fast technology to aid in the characterization of PWS.
Optical coherence tomography (OCT) has been used for non-contact, high-resolution imaging of diseases in clinical practice. Especially, optical coherence tomography angiography (OCTA) has played an important role in imaging of vasculature in vivo [9–11]. To extract the vessel related information, accurate segmentation of OCTA images is the fundamental for effective calculation. In this context, we recently developed the inverse signal-to-noise ratio (iSNR)-decorrelation (D) OCTA (ID-OCTA) which was able to effectively suppress noise and contributed to a proper segmentation [12]. Further, we proposed the binary image similarity (BISIM) method in the ID space, termed ID-BISIM [13], which realized automatic 3D adaptive blood vessel segmentation with a high sensitivity and specificity. However, OCTA technology still has some problems, for example, the presence of the tail shadow with elongated blood vessels in depth direction [14].
To achieve a better understanding of vessel related diseases for improved diagnosis and treatment, it is important to accurately resolve the structure of these blood vessels. However, even as 3D vessel images become readily accessible, quantitative analysis algorithms for their organization have mainly remained limited to analysis of 2D images, with only a few notable exceptions [15,16]. For example, Borrelli et al. collected 3D OCTA data from patients with type 1 diabetes and no signs of diabetic retinopathy diabetes, and developed methods to obtain the 3D vascular volume and 3D perfusion density [16]. Even though, some critical features of blood vessels, such as 3D orientations, alignment and the crimping level, have not yet been fully understood, especially at a voxel-wise basis. In recent years, we have focused on resolving 3D organization and morphology of fiber-like structures, mainly including blood vessels, collagen fibers and elastin fibers, and developed a series of quantitative metrics at a voxel-wise basis (pixel-wise for 2D case), such as orientation [17], alignment [18], waviness [19], local density [20] and thickness [21]. These highly-informative metrics have been successfully applied to gaining a better understanding and enabling highly-accurate diagnosis of diseases including breast cancer, peritoneal metastasis, brain injury, osteoarthritis and stroke [18–23].
In this study, we developed an imaging and analysis system for a better understanding and accurate diagnosis of PWS. Specifically, we acquired the images of blood vessels in vivo using ID-OCTA clinically, and used the mean-subtraction method to correct the shadowing at the depth dimension. We proposed the multi-parametric platform at a voxel-wise basis by integrating the developed parameters of vessel morphology and organization for the characterization of PWS vessel images in a 3D manner. This system enabled label-free, non-invasive vessel image acquisition, and provided insights into vessel alterations from complementary aspects, potentially leading to improvements in diagnosis and treatment of PWS.
2. Methods
2.1 Subject population
This study was approved by the Ethics Committee of the Chinese People’s Liberation Army General Hospital (No. 0404). We recruited 10 patients aged 5-15 years old with PWS on the face, and obtained informed consent from their guardians.
2.2 ID-OCTA image acquisition
The skin imaging was performed with a lab-built spectral domain OCT (SD-OCT) system (Fig. 1). This system used a broadband super luminescent diode (Thorlabs, Newton, NJ, USA, SLD1325) as the light source, which had a central wavelength of 1325 nm and a full width at half maximum bandwidth of 100 nm. The laser power on the skin was measured to be 1 mW, which satisfied the ANSI safety standard. The lateral resolution of the system was tested to be 10 µm, and the axial resolution was 7.6 µm. The scanning range used in this study was 2.5 × 2.5 × 2.5 mm (x × y × z), including 240 × 240 × 412 pixels, and the total acquisition time for one field was about 8.3 s. Data were undersampled so as to reduce the imaging time as much as possible for the young patients on the premise of being able to distinguish the majority of blood vessels. After the cheek skin of each subject was cleaned, OCT images were acquired from the lesion skin and the corresponding normal skin (symmetrical parts of skin lesions on the cheek) from each subject using the hand-held probe. It should be noted that breathing fluctuations in the measurement process would have an impact [11]. Therefore, two neighboring fields were imaged from each site of each subject’s cheek with their quantification results averaged to reduce the fluctuation of measurements. Finally, a total of 36 fields were obtained in our study, including 18 fields from the lesion skin and 18 ones from the normal skin.
Fig. 1. The schematic diagram of the imaging system. Left: the schematic showing the PWS imaging. Right: the detailed layout of the OCTA system. SLD: super luminescent diode; PC: polarization controller; CL: collimate lens; FL: focus lens; OL: objective lens.
A signal-to-noise (SNR) adaptive OCTA algorithm was used to identify dynamic flow, termed as ID-OCTA. According to the asymptotic relationship between the inverse SNR (iSNR) and decorrelation (D) derived based on the multivariate time series (MVTS) model, the classification line was expressed as [12]:
where $E(D)$ and $\sigma$ represented the mean value and the standard deviation (std) of decorrelation, $G \approx 1.5$ was a coefficient of variance (CoV) parameter, and $N = x \times y \times z \times t$ was the spatio-temporal kernel size, which was used for decorrelation calculation. As a tradeoff between the spatial resolution and classification performance, a practical spatio-temporal kernel of $3 \times 3 \times 3 \times 5(x \times y \times z \times t)$ was used in human skin data.As shown in Fig. 2(a), human skin structural images were generated based on the intensity of OCT signals. Decorrelation images [Fig. 2(b)] were then created by calculating the changes in OCT signals during continuous cross-sections. However, it was very difficult and not accurate enough to segment dynamic flow signals simply with a fixed threshold in decorrelation dimension. As shown in Fig. 2(b), the deep static regions with low SNR levels, also had high decorrelation values, resulting in noise-induced dynamic artifacts. To address this issue, many adaptive thresholding methods were proposed for processing OCTA images [12,13,24,25]. Based on the asymptotic relationship between decorrelation coefficient D(x, y, z) of the static signal and the inverse signal-to-noise ratio (iSNR), we created the SNR-adaptive OCT angiogram [Fig. 2(c)]. The pixels in the cross-sections were plotted in the ID space, and downsampled for better display [Fig. 2(d)]. The defined classification line $\mathop D\nolimits_c$ [see the red line in Fig. 2(d), determined by Eq. (1)] separated the ID space into 2 parts: 1) static/noise signal part, such as static tissue and air (below the red line); 2) dynamic signal part, such as dynamic flow (above the red line) [12]. To ease understanding, dynamic signal, static signal and noise are marked in Fig. 2(c) by the arrow. In this way, the 3D ID-OCTA image and corresponding enface projection were obtained accordingly [Fig. 2(e)].
Fig. 2. ID-OCTA of human skin using an SNR-adaptive method. (a) Cross section of OCT intensity map. (b) Decorrelation map. (c) Dynamic flow signal (red area) identified with the classification line (determined by Eq. (1)). Dynamic signal, static signal and noise are marked by the arrow. (d) ID space mapping. The red line is the classification line. (e) Color-coded 3D ID-OCTA image and enface projection of ID-OCTA image. Scale bar: 0.5 mm.
2.3 Assessment of ID-OCTA image quality
To assess the quality of images obtained from ID-OCTA, contrast-to-noise ratio (CNR) was calculated as follows [26]:
where ns and nb referred to the mean values of decorrelation in signal and noise regions, and σb demoted the standard variance in the noise region.To reduce the tail artifacts below the large blood vessel, we used a mean-subtraction method that calculated a weighted mean of all pixels in each A-line [14]. Mathematically, this method was expressed as:
where $OCT{A_{DS}}(i)$ and $OCTA(i)$ represented the ith A-lines of the de-shadowed and original image, respectively; $OCTA(i,k)$ was the magnitude at the kth pixel point of the ith A-line of the original image; k was equal to the number of all pixel points of the A-line; ω was a weighted average, which was set between 1 and 3. A comparison of the image before and after the correction is shown in Figure S1 of Supplement 1.2.4 Overview of multi-parametric analysis
To quantitatively assess the morphology and organization of blood vessels and to tell the difference between the normal and the lesion skin, we developed the multi-parametric analysis platform by integrating parameters including orientation, directional variance, waviness, thickness and local coverage. These parameters were generated based on both enface maximum intensity projection in a 2D manner and truly 3D vessel image, and the sensitivity acquired from 2D and 3D analysis was then compared.
For the maximum projection of 3D vessel image which reduced the 3D image to the 2D image, only one azimuthal angle $\theta$ was needed to define the vessel orientation, and the 2D directional variance $\overline {{D_{2D}}}$ was first calculated as follows [27]:
For 3D case, we developed the 3D weighted orientation vector summation algorithm to resolve 3D orientations of vessels [17]. Further, the 3D directional variance was proposed to describe the degree of alignment in 3D space [18]. In 3D case, an azimuth angle $\theta$ (ranging from 0°-180°) and a polar angle $\varphi$ (ranging from 0°-180°) were used to define a certain orientation [ Fig. 3(a)]. However, since the calculation of the polar angle $\varphi$ was not straightforward, we defined another two angles which were symmetrical to the azimuthal angle $\theta$, named $\beta$ and $\gamma$[Fig. 3(a)]. $\beta$ was defined as the angle between the projection of the vessel in the zx plane and the x axis, and $\gamma$ was the angle between the projection in the yz plane and the −y axis. These two angles were related to the polar angle $\varphi$ as follows:
Then the 3D directional variance $\overline {{D_{3D}}}$ was defined as:
Fig. 3. Multi-parametric analysis of morphological and organizational features of PWS vessel images. (a) Definition of angles and representative raw ID-OCTA image, along with corresponding $\theta$ and $\varphi$ orientation map, variance map, waviness map, local coverage map and thickness map, in both 3D and 2D formats. Scale bar: 0.5 mm. (b) Distribution histograms for each parameter.
Besides directional variance, we obtained other morphology and organization characteristics as well. The waviness measurement described the degree of curvature for vessel structures, and its value ranged from 0 to 1 [19]. A value closer to 1 corresponded to a higher level of crimping [Fig. 3(a)]. To calculate 2D waviness, we obtained the $\theta$ orientation distribution of the vessel structure and performed cross-correlation calculations on its adjacent windows. In this way, the difference between the central vessel pixel and the other pixels within the window could be obtained, which was then normalized to acquire the waviness value [19]. To extend the waviness quantification from 2D to 3D, we acquired the orientation distribution of all the three azimuthal angles, i.e., $\theta$, $\beta$ and $\gamma$, with the final 3D waviness extracted by averaging the waviness obtained from each of them [19]. The custom code for the waviness calculation is available at: https://github.com/ShuhaoQian/Waviness_calculation_code.
The thickness metric described the pixel-wise diameter of the vessel structure [21]. For the binarized image, we extracted the minimum distance d from each point identified as blood vessels to the background, and then performed the adaptive distance transmission so that pixels perpendicular to the vessel at each distinct location had approximately the same distance value. After that, to mimic the continuous change in thickness of blood vessels, we applied the adaptive correlation operator to smooth the distance transmission image, and finally acquired the thickness map through pseudo-color coding [Fig. 3(a)] [21].
The local coverage metric c described the proportion of vessel structure pixels over the total number of pixels in the window [20], shown as follows:
where A(x, y) represented the binarized image within the window, and n2 indicated the total pixel number in the window [Fig. 3(a)] [20]. It should be noted that the local coverage here was a relative measurement, and it was necessary to take the same window size for the normal and the lesion skin for comparison. In contrast to directional variance and waviness metrics which were distinct for 3D and 2D analysis, the above-mentioned calculation methods for the other two metrics, i.e., thickness and local coverage, were applicable to both 2D and 3D cases.The demonstration of the multi-parametric analysis using a representative PWS image stack is shown in Fig. 3(a) by providing maps of each metric for both 3D images and their 2D projections. Then corresponding distribution histograms for these parameters were acquired accordingly [Fig. 3(b)]. A relatively uniform distribution of $\theta$ orientation indicated that these vessels were not inclined to a certain direction. It is worth mentioning that here the orientation referred to the morphological characteristic of blood vessels, while not that of the blood flow.
2.5 Statistical analysis
Student’s t-test was carried out to assess significant differences between the normal and the lesion skin for both 3D and 2D analysis. Results were considered significant at p < 0.05. The box plots were drawn using JMP 13.
A t-Distributed Stochastic Neighbor Embedding (t-SNE) classification model was constructed using three parameters for dimension reduction and demonstration of data. The area under the curve (AUC) value of the receiver operating curve (ROC) was obtained using IBM SPSS Statistics. A typical linear discriminant analysis was performed to obtain the original classification accuracy (OCA) and the cross-validated classification accuracy (CVCA).
3. Results
3.1 Performance of ID-OCTA
A representative maximum intensity projection of the vessel image obtained from the ID-OCTA system is shown in Fig. 4(a). Two regions of interest (ROIs) were zoomed in, where blood vessels were clearly resolved. To assess the image quality, we generated histograms of vascular signal and background noise of this image [Fig. 4(b)], and obtained a CNR value of 4.67, which indicated that the ID-OCTA method used in this study was able to effectively suppress the noise level and resulted in blood vessel images with a high quality. The superior performance of the imaging system was the fundamental for subsequent vessel quantifications.
Fig. 4. Performance of ID-OCTA imaging. (a) A representative maximum intensity projection of the vessel image. Two regions of interest (ROIs) are zoomed in. Scale bar: 1 mm. (b) Histograms of vascular signal and background noise corresponding to this representative image. The CNR value is indicated in the graph.
3.2 Multi-parametric characterization of skin vascular images
The multi-parametric analysis was then performed on blood vessel images. Four representative maps were shown as examples for each parameter, including variance [ Fig. 5(a)], waviness [Fig. 5(b)], thickness [Fig. 5(c)] and local coverage [Fig. 5(d)]. As mentioned above, the variance metric quantified the alignment of vessels, and PWS lesions led to redder hues in variance maps, corresponding to a degradation in vessel alignment [Fig. 5(a)]. The waviness metric characterized the crimping level of vessels, and no obvious differences were observed between the lesion and the normal skin [Fig. 5(b)]. In PWS skin tissues, we found that the thickness maps exhibited redder hues, revealing an increase in vessel diameter [Fig. 5(c)]. Accordingly, the thickening of blood vessels in PWS tissues might lead to an elevated level of local coverage, which quantified the density of vessels at local regions [Fig. 5(d)].
Fig. 5. Multi-parametric analysis of blood vessel images from the PWS lesion and the normal skin. (a-d) Multi-parametric maps from 4 representative fields, including variance (a), waviness (b), thickness (c) and local coverage (d) maps. Scale bar: 0.5 mm. (e-h) Boxplots showing statistical analysis results for variance (e), waviness (f), thickness (g) and local coverage (h), using both 2D and 3D analysis methods. *, p < 0.05; **, p < 0.01; ***, p < 0.001.
Then these qualitative observations were further validated by quantitative analysis. In order to compare sensitivity obtained from 2D and 3D approaches, we performed analysis on both 3D stack images and 2D projections, from totally 36 skin vessel fields (including 18 normal and 18 lesions) of 10 patients. As expected, there was a degradation in vessel alignment (i.e., a higher level of variance) as quantified from truly 3D analysis [Fig. 5(e)]. Interestingly, this significant difference could be resolved from 3D analysis only, while not from 2D analysis, although the same trends were observed [Fig. 5(e)]. Consistent with observations from waviness maps, there were no significant differences in the vessel crimping level between the lesion and normal skin, either from 2D or 3D analysis, indicating that this type of vascular malformation did not intensely alter the vessel curvature [Fig. 5(f)]. Moreover, we obtained a significant increase in vessel thickness [Fig. 5(g)] and local coverage [Fig. 5(h)] as a result of PWS lesions from both 2D and 3D analyses, supporting hues observed from corresponding maps. It was interesting to see that the thickness quantification was higher from 2D over 3D analysis, which was not surprising because 2D analysis was performed on the maximum projection of 3D vessel images and the sum of the signal along the depth dimension might be a possible reason.
3.3 Classification of PWS and normal skin based on multi-parametric analysis
Inspired by the analysis results which indicated few differences in vessel waviness induced from PWS lesions, we proposed a strategy which integrated sensitive parameters for PWS including variance, thickness and local coverage for classification of the PWS lesion and the normal skin. The scatter plot to visualize data clustering is shown in Fig. 6(a), and detailed classification results [Fig. 6(b)] and corresponding receiver operating curve (ROC) [Fig. 6(c)] further highlighted the strong discriminative power of the selected parameter combination. Specifically, we obtained 91.7% of the original classification accuracy (OCA) and 83.3% of the cross-validated classification accuracy (CVCA). It was worth mentioning that this selected multi-parametric model led to the area under the curve (AUC) value of 0.978, strongly confirming the sensitivity of this approach. Further, we visualized the clustering of data using the t-Distributed Stochastic Neighbor Embedding (t-SNE), which was a statistical dimension-reduction technique that was particularly well suited for the visualization of high-dimensional datasets [Fig. 6(d), the left most graph] [28]. In addition, we obtained t-SNE gradient maps of different metrics to show numerical variations across normal and PWS samples [Fig. 6(d), the right three graphs]. Scatters with similar hues were grouped, indicating that these selected metrics were sensitive to the identification of skin lesions. Corresponding 2D analysis results are shown in Figure S3, which led to 77.8% in OCA and 72.2% in CVCA. The 3D vs. 2D comparison indicated that 3D analysis provided more accurate descriptions of the morphological and organizational changes in blood vessels resulting from PWS skin lesions.
Fig. 6. Classification of PWS and normal skin vessels based on 3D analysis. (a) 3D scatter plot showing multi-parametric analysis-based discrimination of PWS and normal skin samples. (b) Classification results of PWS and normal samples. (c) ROC obtained by combining three parameters, with AUC indicated in the graph. (d) t-SNE classification model (the first graph) and corresponding maps to show the gradient of each parameter across PWS and normal skin vessels.
4. Discussion
Port wine stain (PWS) is a type of birthmark caused by abnormal blood vessels under the skin, but its exact etiology is currently unknown [8,29]. It usually appears as a red or purple patch that darkens and thickens over time, which can severely affect people’s physical and emotional well-being. Therefore, there is an urgent need for an imaging technique which is able to acquire blood vessels images from patients in vivo in a non-invasive, label-free manner for the subsequent analysis. In 2010, Zhao et al. [30] used conventional OCT to measure vessel diameter and depth in PWS. In recent years, OCTA has been used as a typical 3D visualization method to observe PWS vessels [31,32]. Besides ID-OCTA proposed in this study, other OCTA processing algorithms also led to nice identification of blood vessels, such as optical microangiography (OMAG) algorithm [33–35]. The main difference between them was that OMAG algorithm obtained blood flow information by comparing the interference of OCT signals obtained from several scans at the same position, while ID-OCTA algorithm obtained blood flow information by analyzing the intensity changes of OCT signals. OMAG technology had high sensitivity, while it might also be easily affected by factors such as motion and projection artifacts [33]. ID-OCTA technology could reduce these artifacts to some extent, but required more complex algorithms and higher signal-to-noise ratio. In this study, we used hand-held probe based ID-OCTA to acquire blood vessels images clinically [36–38], which effectively suppressed noise for improved image quality, and resulted in a high CNR of the vessel image. It is worth noting that the tail artifacts of the vessel have always been a problem in OCTA technology. Methods were developed to reduce the tail artifacts [39–41], such as the mean-subtraction method [14], the slab-subtraction (SS) method [42], the reflectance-based projection-resolved (rbPR) OCTA algorithm [43], and the normalized field autocorrelation function-based method [44]. The method we adopted was the mean-subtraction method, which was suitable for large vessels and helped us to get closer to the real 3D architecture of blood vessels. These improvements in image acquisition paved the way for the following quantitative characterizations.
We developed a multi-parametric analysis platform which integrated a series of morphological and organizational features including orientation, alignment, waviness, thickness and local density, and achieved voxel-wise assessments of vessel deformation in PWS lesions. These metrics offered complementary insights into the architecture of blood vessels within skin, and provided a clear picture of the abnormal vessel configuration in PWS. Specifically, we observed a degradation in vessel alignment, and an increase in vessel thickness and density induced by skin lesions. Moreover, it was interesting to find that the crimping level of blood vessels was not affected a lot. One limitation of our analysis method was that it quantified the morphological and organizational characteristics of blood vessels only, while was not able to tell the orientation of blood flow, which would be our future work. The complementarity of these optical metrics also led to a high classification accuracy in identifying PWS lesions, further highlighting this multi-parametric analysis platform as a powerful tool for diagnosis of PWS and assessments of PWS treatment, such as PDT [31].
Our analysis results also validated that 3D analysis led to a higher sensitivity than 2D analysis. As shown in Fig. 5(e), only 3D analysis was able to tell the difference in alignment between the normal and the lesion skin. Moreover, the multi-parametric analysis in a 3D context led to a classification accuracy of ∼90% in identifying PWS lesions, in contrast to ∼75% obtained from the 2D compartment. A possible reason was that 3D analysis stacked up the vessels and assessed the vessel organization in its real architecture, without discarding useful information from both azimuthal and polar angles, while 2D analysis relied on the information from the azimuthal angle only.
5. Conclusion
In this study, we develop an in vivo imaging and automatic analysis system for characterizations of PWS. The OCTA imaging system enables non-invasive image acquisition without any exogenous labeling agents, and the ID-OCTA method leads to superb image quality. The multi-parametric analysis platform developed in this study offers thorough and complementary insights into vascular deformations induced by PWS lesions. Especially, this analysis platform can operate in a 3D format at a voxel-wise basis, potentially leading to advances in analysis sensitivity compared with that from conventional 2D analysis. Overall, this system has potential in contributing to improvements in diagnosis and treatment of PWS.
Funding
National Key Research and Development Program of China (2019YFE0113700, 2017YFA0700501); Natural Science Foundation of Zhejiang Province (LR20F050001, LR19F050002); National Natural Science Foundation of China (62275232, 61905214, 62075189, 62035011, 11974310, 31927801, 61835015, T2293751).
Acknowledgments
We thank Prof. Chen-Yuan Dong at National Taiwan University for insightful discussion and Alibaba Cloud.
Disclosures
The authors declare no conflict 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.
Supplemental document
See Supplement 1 for supporting content.
References
1. J. K. Chen, P. Ghasri, G. Aguilar, A. M. van Drooge, A. Wolkerstorfer, K. M. Kelly, and M. Heger, “An overview of clinical and experimental treatment modalities for port wine stains,” J. Am. Acad. Dermatol. 67(2), 289–304.e29 (2012). [CrossRef]
2. N. Demirel Ogut, “Acquired plantar port-wine stain (Fegeler syndrome) following prolonged standing,” J. Cosmet. Dermatol. 21(10), 5234–5235 (2022). [CrossRef]
3. F. Jiang, J. Shao, L. Chen, N. Yang, J. Liu, and Z. Li, “Influence of Port-wine Stains on Quality of Life of Children and Their Parents,” Acta Derm.-Venereol. 101(8), adv00516 (2021). [CrossRef]
4. L. Liu, R. Mei, Y. Pan, E. Li, Q. Zhao, and X. Jiang, “Successful treatment of hypertrophic and nodular port-wine stains with intralesional 1064 nm Nd: YAG laser,” Dermatol Ther 33, e13925 (2020). [CrossRef]
5. J. Kowalska-Brocka, M. Brocki, S. Uczniak, K. Uczniak, A. Kaszuba, and P. Jurowski, “Sturge-Weber syndrome type II treated with PDL 595 nm laser,” Postepy Dermatol Alergol 1, 63–66 (2015). [CrossRef]
6. L. C. Zhang, J. Yang, Y. B. Huang, Z. Huang, and M. Y. Bi, “Post treatment care in photodynamic therapy (PDT) of large facial port-wine stain (PWS) birthmarks,” Photodiagn. Photodyn. Ther. 36, 102604 (2021). [CrossRef]
7. K. C. Tark, D. H. Lew, and D. W. Lee, “The fate of long-standing port-wine stain and its surgical management,” Plast Reconstr Surg 127(2), 784–791 (2011). [CrossRef]
8. M. Zhang, Q. Wu, T. Lin, L. Guo, Y. Ge, R. Zeng, Y. Yang, H. Rong, G. Jia, Y. Huang, J. Fang, H. Shi, W. Zhao, S. Chen, and P. Cai, “Hematoporphyrin monomethyl ether photodynamic therapy for the treatment of facial port-wine stains resistant to pulsed dye laser,” Photodiagn. Photodyn. Ther. 31, 101820 (2020). [CrossRef]
9. M. Alam, D. Thapa, J. I. Lim, D. Cao, and X. Yao, “Computer-aided classification of sickle cell retinopathy using quantitative features in optical coherence tomography angiography,” Biomed. Opt. Express 8(9), 4206–4216 (2017). [CrossRef]
10. C. Blatter, J. Weingast, A. Alex, B. Grajciar, W. Wieser, W. Drexler, R. Huber, and R. A. Leitgeb, “In situ structural and microangiographic assessment of human skin lesions with high-speed OCT,” Biomed. Opt. Express 3(10), 2636–2646 (2012). [CrossRef]
11. W. Gao, “Quantitative depth-resolved microcirculation imaging with optical coherence tomography angiography (Part IotaIota): Microvascular network imaging,” Microcirculation 25(6), e12376 (2018). [CrossRef]
12. L. Huang, Y. Fu, R. Chen, S. Yang, H. Qiu, X. Wu, S. Zhao, Y. Gu, and P. Li, “SNR-Adaptive OCT Angiography Enabled by Statistical Characterization of Intensity and Decorrelation With Multi-Variate Time Series Model,” IEEE Trans. Med. Imaging 38(11), 2695–2704 (2019). [CrossRef]
13. Y. Zhang, H. Li, T. Cao, R. Chen, H. Qiu, Y. Gu, and P. Li, “Automatic 3D adaptive vessel segmentation based on linear relationship between intensity and complex-decorrelation in optical coherence tomography angiography,” Quant Imaging Med Surg 11(3), 895–906 (2020). [CrossRef]
14. W. J. Choi, B. Paulson, S. Yu, R. K. Wang, and J. K. Kim, “Mean-Subtraction Method for De-shadowing of Tail Artifacts in Cerebral OCTA Images: A Proof of Concept,” Materials 13(9), 2024 (2020). [CrossRef]
15. J. Zhang, Y. C. Qiao, M. S. Sarabi, M. M. Khansari, J. K. Gahm, A. H. Kashani, and Y. G. Shi, “3D Shape Modeling and Analysis of Retinal Microvasculature in OCT-Angiography Images,” IEEE Trans. Med. Imaging 39(5), 1335–1346 (2020). [CrossRef]
16. E. Borrelli, D. Grosso, M. Parravano, E. Costanzo, M. Brambati, C. Vigano, R. Sacconi, L. Querques, A. Pina, D. De Geronimo, F. Bandello, and G. Querques, “Volume rendered 3D OCTA assessment of macular ischemia in patients with type 1 diabetes and without diabetic retinopathy,” Sci. Rep. 11(1), 19793 (2021). [CrossRef]
17. Z. Liu, K. P. Quinn, L. Speroni, L. Arendt, C. Kuperwasser, C. Sonnenschein, A. M. Soto, and I. Georgakoudi, “Rapid three-dimensional quantification of voxel-wise collagen fiber orientation,” Biomed. Opt. Express 6(7), 2294–2310 (2015). [CrossRef]
18. Z. Liu, D. Pouli, D. Sood, A. Sundarakrishnan, C. K. Hui Mingalone, L. M. Arendt, C. Alonzo, K. P. Quinn, C. Kuperwasser, L. Zeng, T. Schnelldorfer, D. L. Kaplan, and I. Georgakoudi, “Automated quantification of three-dimensional organization of fiber-like structures in biological tissues,” Biomaterials 116, 34–47 (2017). [CrossRef]
19. S. Qian, J. Meng, Z. Feng, L. Zhou, S. Jiang, Y. Wang, Z. Ye, S. Zhuo, J. Chen, X. Li, L. Gao, Z. Ding, J. Qian, and Z. Liu, “Mapping Organizational Changes of Fiber-Like Structures in Disease Progression by Multiparametric, Quantitative Imaging,” Laser Photonics Rev. 16, 2100576 (2022). [CrossRef]
20. S. Qian, G. Wang, J. Meng, S. Jiang, L. Zhou, J. Lu, Z. Ding, S. Zhuo, and Z. Liu, “Identification of human ovarian cancer relying on collagen fiber coverage features by quantitative second harmonic generation imaging,” Opt. Express 30(14), 25718–25733 (2022). [CrossRef]
21. J. Meng, Z. Feng, S. Qian, C. Wang, X. Li, L. Gao, Z. Ding, J. Qian, and Z. Liu, “Mapping physiological and pathological functions of cortical vasculature through aggregation-induced emission nanoprobes assisted quantitative, in vivo NIR-II imaging,” Biomater. Adv. 136, 212760 (2022). [CrossRef]
22. C. K. H. Mingalone, Z. Y. Liu, J. M. Hollander, K. D. Garvey, A. L. Gibson, R. E. Banks, M. Zhang, T. E. McAlindon, H. C. Nielsen, I. Georgakoudi, and L. Zeng, “Bioluminescence and second harmonic generation imaging reveal dynamic changes in the inflammatory and collagen landscape in early osteoarthritis,” Lab. Invest. 98(5), 656–669 (2018). [CrossRef]
23. Z. Y. Liu, L. Speroni, K. P. Quinn, C. Alonzo, D. Pouli, Y. Zhang, E. Stuntz, C. Sonnenschein, A. M. Soto, and I. Georgakoudi, “3D organizational mapping of collagen fibers elucidates matrix remodeling in a hormone-sensitive 3D breast tissue model,” Biomaterials 179, 96–108 (2018). [CrossRef]
24. H. Li, K. Liu, T. Cao, L. Yao, Z. Zhang, X. Deng, C. Du, and P. Li, “High performance OCTA enabled by combining features of shape, intensity, and complex decorrelation,” Opt. Lett. 46(2), 368–371 (2021). [CrossRef]
25. S. S. Gao, Y. Jia, L. Liu, M. Zhang, H. L. Takusagawa, J. C. Morrison, and D. Huang, “Compensation for Reflectance Variation in Vessel Density Quantification by Optical Coherence Tomography Angiography,” Invest. Ophthalmol. Vis. Sci. 57(10), 4485–4492 (2016). [CrossRef]
26. Z. Y. Liu, S. H. Ma, Y. H. Ji, L. Liu, Z. X. Hu, J. H. Guo, H. Ma, and Y. H. He, “Quasi-Confocal, Multichannel Parallel Scan Hyperspectral Fluorescence Imaging Method Optimized for Analysis of Multicolor Microarrays,” Anal. Chem. 82(18), 7752–7757 (2010). [CrossRef]
27. Kanti V. Mardia and Peter E. Jupp, Directional Statistics (John Wiley & Sons, Inc.), p. 1999.
28. L. van der Maaten and G. Hinton, “Visualizing Data using t-SNE,” J Mach Learn Res 9, 2579–2605 (2008).
29. W. Tan, J. Wang, F. Zhou, L. Gao, R. Yin, H. Liu, A. Sukanthanag, G. Wang, M. C. Mihm Jr., D. B. Chen, and J. S. Nelson, “Coexistence of Eph receptor B1 and ephrin B2 in port-wine stain endothelial progenitor cells contributes to clinicopathological vasculature dilatation,” Br J Dermatol 177(6), 1601–1611 (2017). [CrossRef]
30. S. Zhao, Y. Gu, P. Xue, J. Guo, T. Shen, T. Wang, N. Huang, L. Zhang, H. Qiu, Y. Xin, and X. Wei, “Imaging port wine stains by fiber optical coherence tomography,” J. Biomed. Opt. 15(3), 036020 (2010). [CrossRef]
31. Y. D. Liu, D. Zhu, J. J. Xu, Y. Wang, W. Feng, D. F. Chen, Y. Q. Li, H. L. Liu, X. H. Guo, H. X. Qiu, and Y. Gu, “Penetration-enhanced optical coherence tomography angiography with optical clearing agent for clinical evaluation of human skin,” Photodiagn. Photodyn. Ther. 30, 101734 (2020). [CrossRef]
32. J. S. Waibel, J. Holmes, A. Rudnick, D. Woods, and K. M. Kelly, “Angiographic optical coherence tomography imaging of hemangiomas and port wine birthmarks,” Lasers Surg. Med. 50(7), 718–726 (2018). [CrossRef]
33. Q. Zhang, A. Zhang, C. S. Lee, A. Y. Lee, K. A. Rezaei, L. Roisman, A. Miller, F. Zheng, G. Gregori, M. K. Durbin, L. An, P. F. Stetson, P. J. Rosenfeld, and R. K. Wang, “Projection artifact removal improves visualization and quantitation of macular neovascularization imaged by optical coherence tomography angiography,” Ophthalmol. Retina 1(2), 124–136 (2017). [CrossRef]
34. W. J. Choi, R. Reif, S. Yousefi, and R. K. Wang, “Improved microcirculation imaging of human skin in vivo using optical microangiography with a correlation mapping mask,” J. Biomed. Opt. 19(3), 036010 (2014). [CrossRef]
35. Y. Huang, Q. Zhang, M. R. Thorell, L. An, M. K. Durbin, M. Laron, U. Sharma, G. Gregori, P. J. Rosenfeld, and R. K. Wang, “Swept-source OCT angiography of the retinal vasculature using intensity differentiation-based optical microangiography algorithms,” Ophthalmic Surg Lasers Imaging Retina 45(5), 382–389 (2014). [CrossRef]
36. 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]
37. R. Mastropasqua, L. Toto, L. Di Antonio, E. Borrelli, A. Senatore, M. Di Nicola, G. Di Martino, M. Ciancaglini, and P. Carpineto, “Optical coherence tomography angiography microvascular findings in macular edema due to central and branch retinal vein occlusions,” Sci. Rep. 7(1), 40763 (2017). [CrossRef]
38. B. Bhaduri, R. M. Nolan, R. L. Shelton, L. A. Pilutti, R. W. Motl, H. E. Moss, J. H. Pula, and S. A. Boppart, “Detection of retinal blood vessel changes in multiple sclerosis with optical coherence tomography,” Biomed. Opt. Express 7(6), 2321–2330 (2016). [CrossRef]
39. M. Zhang, T. S. Hwang, J. P. Campbell, S. T. Bailey, D. J. Wilson, D. Huang, and Y. Jia, “Projection-resolved optical coherence tomographic angiography,” Biomed. Opt. Express 7(3), 816–828 (2016). [CrossRef]
40. Y. Li and J. Tang, “Blood vessel tail artifacts suppression in optical coherence tomography angiography,” Neurophotonics 9(02), 021906 (2022). [CrossRef]
41. A. Zhang, Q. Zhang, and R. K. Wang, “Minimizing projection artifacts for accurate presentation of choroidal neovascularization in OCT micro-angiography,” Biomed. Opt. Express 6(10), 4130–4143 (2015). [CrossRef]
42. Y. Jia, S. T. Bailey, D. J. Wilson, O. Tan, M. L. Klein, C. J. Flaxel, B. Potsaid, J. J. Liu, C. D. Lu, M. F. Kraus, J. G. Fujimoto, and D. Huang, “Quantitative Optical Coherence Tomography Angiography of Choroidal Neovascularization in Age-Related Macular Degeneration,” Ophthalmology 121(7), 1435–1444 (2014). [CrossRef]
43. J. Wang, M. Zhang, T. S. Hwang, S. T. Bailey, D. Huang, D. J. Wilson, and Y. Jia, “Reflectance-based projection-resolved optical coherence tomography angiography,” Biomed. Opt. Express 8(3), 1536–1548 (2017). [CrossRef]
44. J. Tang, S. E. Erdener, S. Sunil, and D. A. Boas, “Normalized field autocorrelation function-based optical coherence tomography three-dimensional angiography,” J. Biomed. Opt. 24(03), 1–8 (2019). [CrossRef]
