Accurate Detection of 3D Choroidal Vasculature Using Swept-Source OCT Volumetric Scans Based on Phansalkar Thresholding

Various eye ailments associated with the posterior segment of the eye, including age-related macular degeneration (AMD) and central serous chorioretinopathy (CSCR), are caused due to dysfunction of the highly vascular choroid layer. It is responsible for supplying oxygen and nutrients to the retinal outer layers and maintaining the thermal equilibrium of the eye. Clinicians hypothesize that detecting minute volumetric structural changes of the choroidal vasculature enables early diagnosis. To this end, recently introduced swept-source optical coherence tomography (SS-OCT) volume scans provide dense high-resolution imaging of the choroid. However, due to intricate structure, manual segmentation of these vessels is not feasible and clinicians seek algorithmic segmentation and attempts made earlier reported limited performance. In response, we propose a method based on adaptive Phansalkar thresholding to accurately detect choroidal vessels in OCT volumes. Specifically, it increases the contrast between vessel and non-vessel regions within each subblock of the B-Scan. On 15 SS-OCT volumes of healthy and diseased subjects, we performed subjective grading-based performance analysis on 2D and 3D vasculatures achieving 92.67% and 94% segmentation accuracy, respectively. Further, the proposed method demonstrated significant improvement over the previously reported method. Finally, we envisage that this method provides ground truth segmentation for training deep learning models.