TIME-RESOLVED LASER SPECKLE CONTRAST IMAGING FOR DEPTH-SENSITIVE MONITORING OF CEREBRAL HEMODYNAMICS

Cerebral blood flow (CBF) is essential for maintaining brain viability by delivering oxygen and nutrients and removing metabolic waste. Disruptions in CBF are central to the pathophysiology of many neurological disorders, including traumatic brain injury, stroke, hydrocephalus, and neonatal brain injury. However, existing methods for monitoring cerebral hemodynamics remain limited by fundamental tradeoffs, including poor spatiotemporal resolution, low penetration depth, reliance on contact-based measurements, and time-consuming image reconstruction. Conventional camera-based optical imaging methods such as laser speckle contrast imaging (LSCI) provide high-resolution, wide-field flow measurements but are restricted to superficial tissue. In contrast, diffuse optical methods, including near-infrared spectroscopy (NIRS) and diffuse correlation spectroscopy/tomography (DCS/DCT), can probe deeper tissue but suffer from sparse spatial sampling, limited resolution and field-of-view, partial volume effects, and complex reconstruction. This dissertation addresses these limitations through the invention, development, physiological application, and computational enhancement of time-resolved laser speckle contrast imaging (TR-LSCI), a novel, noncontact, cost-efficient, and depth-resolved optical imaging platform for monitoring cerebral hemodynamics. In the first study, a TR-LSCI system was developed and validated for depth-sensitive, wide-field imaging of CBF. The system integrates pulsed, wide-field near-infrared illumination with a synchronized time-gated single-photon avalanche diode (SPAD) camera to differentiate photon path lengths and generate depth-resolved two-dimensional blood flow maps without tomographic reconstruction. Parallelized convolution-based algorithms reduced reconstruction time from tens of minutes to seconds per gated map. Validation in 3D-printed head-simulating phantoms and in-vivo rodent models demonstrated that TR-LSCI can continuously map CBF changes with depth sensitivity under physiological perturbations. The system captured global increases in CBF during CO2 inhalation and regional and global decreases during transient common carotid artery (CCA) ligations in rodents. It also visualized cerebrovascular structures with spatial resolutions ranging from tens to hundreds of micrometers on rodent head surfaces to approximately 1-2 millimeters in deeper brain regions. These results establish TR-LSCI as a fast, noncontact, and depth-resolved platform for CBF imaging. In the second study, the TR-LSCI platform was applied to investigate cerebral autoregulation (CA) during controlled intracranial pressure (ICP) elevation. The system was upgraded to a higher performance SPAD512² camera and redesigned into a portable, fiber-coupled configuration to improve imaging sensitivity, acquisition speed, and translational flexibility. To improve the accuracy of CBF estimation under low photon conditions, a noise-corrected speckle analysis framework was incorporated to account for dark noise and shot noise contributions before calculating the blood flow index. Using this optimized system, wide-field microvascular CBF imaging was combined with synchronized invasive measurements of ICP and arterial blood pressure (ABP) in adult rats and neonatal piglets. TR-LSCI resolved spatially heterogeneous and pulsatile CBF dynamics during graded ICP loading. Heart-rate estimates derived from TR-LSCI were statistically equivalent to those obtained from ICP and ABP, confirming preservation of physiologically meaningful pulsatile information. Multimodal analysis revealed reproducible, phase-dependent transitions in cerebral autoregulation, progressing from preserved autoregulation to ABP-driven compensation and ultimately to ICP-constrained CBF suppression. Notably, CBF dynamics alone exhibited distinct signatures corresponding to these phases, indicating that continuous optical monitoring of CBF may provide insight into autoregulatory status without invasive blood pressure measurements. In the third study, a key limitation of TR-LSCI, photon starvation and diffusive noise at deeper gates was addressed using deep learning. A multiscale latent diffusion model, LTDiff++, was developed to suppress photon diffusion noise while preserving vascular structures. Trained on head-simulating phantom data and neonatal rat imaging datasets, LTDiff++ significantly improved image quality at deeper gates, with consistent increases in structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR). Importantly, the model enabled high-quality imaging using only five averaged frames instead of approximately 100 frames conventionally required, reducing acquisition time by nearly 20-fold. These results demonstrate that integrating latent diffusion modeling with TR-LSCI substantially enhances image robustness, temporal efficiency, and practical applicability in photon-limited conditions. Overall, this dissertation establishes TR-LSCI as an optical neuroimaging platform that combines noncontact operation, high spatial resolution, wide-field coverage, depth sensitivity, and rapid imaging for monitoring cerebral hemodynamics. Through system development, physiological validation during intracranial pressure perturbation, and deep learning-based enhancement, this work advances the capability for continuous monitoring of microvascular CBF and CA, and provides a strong foundation for future translational applications in neuroscience research and neonatal care.