AI-integrated imaging processing platforms for three-dimensional macular hemodynamic analysis

The human retina is a complex and light-sensitive tissue in the eye for the perception of visual information, serving as the crucial interface between incoming optical stimuli and the brain’s visual processing pathways. As one of the highest oxygen consumption organs in human body, regulated blood flow control in the retinal microcirculation is essential to satisfy local metabolic demands, making it a potential biomarker for pathology diagnosis. Optical Coherence Tomography (OCT) is a non-invasive in vivo imaging modality that generates high-resolution, three-dimensional images and is widely used in ophthalmology. OCT angiography (OCTA), a functional extension of OCT, enables structural visualization of retinal vasculature by detecting motion contrast from repeated OCT scans. However, OCTA is limited to static imaging of vascular structures and does not capture blood flow dynamics. Furthermore, OCT/OCTA imaging is susceptible to inevitable speckle noise introduced by the low-coherence light, which can obscure structural details and reduce diagnostic accuracy. This thesis focuses on the development of software-based approaches to overcome the existing limitations while maintaining greater clinical feasibility. The approach described in the thesis provides a novel processing protocol to quantify the retinal capillary perfusion variability using consecutive OCTA images obtained by a commercial system. In addition, a deep learning-based architecture is introduced to reconstruct high signal-to-noise ratio images without compromising fine structural information, using a custom-built dual-spectrometer spectral-domain OCT system.