Fig 1.
Labelled image of OCTAVA graphical user interface.
Colored boxes indicate important user controls for optimization of image processing. Red box: the user can modify the image by down sampling or selecting a subregion of the full image for faster processing. Green box: median filter controls. Blue box: Frangi filter controls. Yellow box: vascular analysis controls. Purple box: batch processing controls.
Fig 2.
Flowchart of software workflow.
(a) and example images representing each step in the workflow (b-g). Colored borders in (b-g) correspond to the step with the matching-colored arrow or box in (a). OCTA MIP image (b) is loaded into OCTAVA. Images are first pre-processed in MATLAB to remove noise and enhance the signal intensity of vessel-like structures using a Frangi filter (c). Images are then segmented using the chosen segmentation algorithm to generate a binary mask (d). The image is then sent to ImageJ, where the network is skeletonized (e). The thickness and interconnectivity are measured (f), and the ROIs and network elements are identified and classified based on their thickness and interconnectivity. An overlay image is generated (g) which helps the user confirm that an accurate map of the network architecture was measured. Colors in (f) correspond to the local vessel thickness. Colors in (g) correspond to different architectural components of the network including segments (yellow lines), branches (green lines), mesh regions (light blue lines), isolated elements (dark blue lines), and nodes (red and blue circles). The insets in (g) allow a closer examination of the identified network elements. Finally, the outputs of the ImageJ analysis are sent to MATLAB, which generates and compiles metrics of the network. The large images in (b-g) are 5 mm × 5 mm. The insets in (g) are 1.6 mm × 1.6 mm.
Fig 3.
Visual representation of metrics.
Table 1.
Microvascular network architecture quantitative metrics.
Fig 4.
Comparison of five segmentation algorithms for three images with different apparent VAD.
Top: binarized OCTA MIP images. Bottom: quantitative analysis. All images are OCTA MIP images with dimensions of 5 mm × 5 mm. VAD: vessel area density; CF: connectivity factor.
Fig 5.
Validation of image processing steps and metric results using images of a microfluidic device.
(a-e) and a simulated OCTA MIP image (f-j). The composite image (a) is 4.32 mm × 4.32 mm with a pixel size of 4 μm. The binary mask (b) and overlay image (c) demonstrate the accuracy of the segmentation algorithm. The thickness map (d) demonstrates accurate measurement of the channel diameters throughout the image. The color bars in (d) and (i) indicate vessel diameter in μm. The green areas in the histogram indicate overlap between the blue and yellow bars. The same analysis was repeated for the simulated OCTA MIP image, which was assumed to be 10 mm × 10 mm with a pixel size of 9.3 μm. The overlay images (c, i) include a color-coded notation indicating different structures within the network: segments (yellow); branches (green); nodes (pink and blue circles); mesh regions (blue). (e) and (j) show the measured distribution of diameters in the microfluidic image and simulated OCTA MIP image, respectively. The larger channels in the microfluidic device are 300 μm and the smaller channels are 50 μm.
Table 2.
Comparison between OCTAVA and manual calculation of metrics.
Fig 6.
Boxplots showing the repeatability of various vascular metrics obtained from selected skin datasets.
Black lines (limits on the boxes) represent the range of values obtained; red lines indicate the median value; blue boxes indicate the 25th and 75th percentiles. FD: fractal dimension.
Table 3.
Intrasession repeatability of vascular metrics obtained using OCTAVA.
Fig 7.
OCTAVA applied to OCTA acquired with three different third-party instruments and imaging targets.
Top row: OCTA MIP image of human forearm acquired using a lab-built PS-OCT system. The image is 6.8 mm × 6.8 mm, and the MIP spans an axial range of 600 μm. Image reprinted with permission from [56]. Middle row: OCTA MIP image of mouse ear skin acquired with Telesto OCT system. The image is 2.5 mm × 2.5 mm and the axial span of the MIP is 1 mm. Reprinted with permission from [57]. Copyright (2021) American Chemical Society. Bottom row: OCTA MIP image of human retina acquired with RTVue XR Avanti. The image is 3 mm × 3 mm, and the axial span of the MIP is about 20 μm. Reprinted with permission from [58]. First column contains the original OCTA image; remaining columns show the evaluation steps using OCTAVA (left to right, the binary mask, overlay of skeleton, and map of vessel diameters). In all cases, the binary mask was generated using the Frangi filter (σmax = 8 for top row and σmax = 6 for middle and bottom rows) and fuzzy thresholding segmentation.
Fig 8.
Motion artifacts, poor-quality images, and manual curation.
Images with significant motion artifact (a) must be removed from the analysis since the segmentation algorithm cannot distinguish between the vessels and motion artifacts; the resultant segmented OCTA MIP (b) does not accurately represent the microvascular architecture. For images with minimal motion artifacts, such as the images in (c) and (d), the individual motion artifacts can be removed manually. (e) and (f) show insets of (c) and (d), respectively, demonstrating the removal of the motion artifact using manual curation. Depending on the VAD, individual motion artifacts likely have little impact on the generated metrics, as shown in the histogram of diameters (g) and the table of generated metrics with and without manual curation (h). Note that in (g), the columns representing the results with manual curation are always equal in height or shorter than the automatically generated values since no vessels were added by manual curation in this instance. Square OCTA MIP images in (a-d) are 5 mm × 5 mm and comprise an axial range of 500 μm. The images in (e) and (f) are 0.65 mm × 5 mm.