Three-Dimensional Curved Workflow-Based Optical Coherence Tomography Angiography for Enhancing Atopic Dermatitis Theranostics

Abstract

Optical coherence tomography angiography (OCTA) is a major advancement in imaging, offering high-resolution microvascular volumetric images crucial for diagnosing and studying dermatological diseases. However, current data analysis and clinical evaluation criteria primarily rely on 2-dimensional (2D) imaging results, resulting in imprecise diagnoses due to the substantial loss of 3D curved structures and microvascular details. To address this issue, we propose a high-fidelity 3D curved processing workflow that integrates an artificial neural network (ANN) with a 3D denoising algorithm based on the curvelet transform and optimal orientation flow (OOF). This innovative workflow enables precise 3D segmentation and accurate quantification of dermal layer microvasculature in atopic dermatitis (AD) in vivo. Furthermore, the use of 3D multiparametric microvasculature quantitative metrics establishes a robust framework for assessing the efficacy of AD treatments in 3D images. Our study results demonstrate that skin structure imaging and the dynamic evolution of 3D microvasculature align with observed pathological changes. Compared to traditional 2D analysis, the maximum variation rate of 3D curved multiparametric information is approximately 10%. Consequently, our research marks a significant advancement in the accurate quantification of microvasculature in AD development and theranostics, paving the way for the clinical application of OCTA in dermatology.

Fig. 1.

Schematic of the proposed processing and analyzing framework. (A) Detailed experimental flow, data processing, and analysis process. (B) 3D volume processing method in detail. (C) Framework quantitative parameters extracted from the processed 3D volume.

Fig. 2.

(A) Schematic diagram of OCT system. (B) Photograph of the OCT imaging of the mouse ear. (C) Schematic of imaging of mouse ear. (D) OCT XY direction scanning. (E) Different ways of OCT and OCTA on image reconstruction. (F) Skin layer’s structure diagram. (F and J) Skin layer’s structure diagram and modeling and treatment strategies, respectively. (G and K) White light and OCT results of a mouse ear before and after modeling. (H and L) Layer structure of skin before and after AD image as the red dotted line in (I) and (M). Scale bar, 2 mm. (I) and (M) OCT en face image of normal skin and AD, according to (G) and (K), respectively. Scale bar, 2 mm.

Fig. 3.

(A) An overview of skin layer segmentation for 3D OCT images. (B) U-net structure. (C, F, and G) Original B-scans, ground truth, and prediction from top to bottom, and 3 border lines are marked in different colors: the background–epidermal interface (blue line, AB1), the epidermal–dermal junction (red line, AB2), and the dermal–background interface (green line, AB3). (D) The bottom indicates a schematic of quantitative data calculation, whereas the below image indicates quantitative data results. (H) Loss value of training level. (I) Accuracy rate of segmentation with training level. (E) 3D skin layer segmentation results, showing the 3D skin structure volume (i), dermal 3D mask (ii), and 3D skin layer structure volume segmented (iii), respectively. Scale bar, 1 mm (C, F, and G).

Fig. 4.

(A) Original 3D OCTA volume. (E) 3D OCTA image after curvelet denoising. (I) 3D OCTA image after OFF vessel enhancement. (B, F, and J) B-scan images along the yellow dotted lines (A′), (E′), and (I′) in (A), (E), and (I), respectively. The white dotted circles indicate blood vessel signals, while the red arrows highlight striped noise areas. (C, G, and K) En face images of the original, curvelet-denoised, and OFF-enhanced 3D OCTA images, respectively. (D, H, and L) 1D signals extracted along the yellow dotted lines in (D′), (H′), and (K′), corresponding to ROI-1, ROI-2, and ROI-3 in (C), (J), and (K). Red arrows (1 to 3) indicate vascular signal areas, green arrows (4 to 6) highlight background signal areas, and white dotted circles (7 and 8) mark regions affected by artifacts. (M and N) Rendered 3D microvascular volumes of normal skin and AD.

Fig. 5.

(A) Image data obtained by different processing methods. (B, C, E, and F) Statistical diagram of blood vessel parameters of VA, VAD, VS, and VSD, respectively. (D and G to I) Average value and the variance value of dermal and epidermal layer thickness change. (J) Reflection spectrum of mouse ear at different days. AR, average rate.