TRD3: Endoscopic and Probe-based Coherence Imaging

点击翻译按钮获取中文摘要

Project Summary TRD 3 The goal of this TRD project is to enhance the power and functionality of endoscopic and probe-based OCT. The small form factor of fiber-optic OCT probes affords the capacity to reach remote organs of the human body, enabling OCT to be routinely used for clinical investigation of the coronary arteries, the gastrointestinal tract, and the lung. However, many strategies to improve image contrast through advanced OCT signal collection and processing are incompatible with the spatial and practical constraints of probe-based OCT. This impairs diagnostic performance and feedback to guide interventions. The focus of TRD 3 is to address some of these limitations. OCT derives image contrast from variations in the tissue’s backscattering properties, but subtle differences in the scattering properties can be difficult to identify because the signal from subsurface microstructure adds up coherently, resulting in speckle. Polarization offers a complementary endogenous contrast mechanism that can afford contrast between tissues that are indiscernible in OCT’s backscattering signal. Many tissues with a fibrillar architecture exhibit birefringence and delay light depending on the alignment of its polarization state with the fibrillar tissue components. Specific Aim 1 capitalizes on tissue’s intrinsic birefringence to measure the orientation of fibrillar tissue elements in all three spatial dimensions through fiber-optic imaging probes. This is specifically relevant for imaging birefringent white matter tracts during stereotactic neurosurgery in the brain. Imaging probes containing two imaging channels at distinct illumination angles and interfaced through a multi-channel motor drive unit will be fabricated. Algorithms that leverage the multiple imaging angles and observe additional continuity constraints will be developed to reconstruct 3D vectorial birefringence. Visualizing the 3D orientation of axonal tracts surrounding an intracranial probe will enable microscopic guidance of stereotactic procedures, such as the implantation of stimulation electrodes for deep brain stimulation. Specific Aim 2 responds to the persistent challenge of speckle in OCT by leveraging machine learning to encapsulate the physical meaning of hardware-based speckle suppression into a trained algorithm. A novel method to generate ground truth speckle-suppressed tomograms using sample tilting for angular compounding will be developed to enable supervised training of a deep neural network. The specific challenge of deploying the trained algorithm to new imaging systems will be addressed by developing both a supervised and an unsupervised method for domain adaptation. Improved image contrast and speckle suppression are critical for interpretation of many tissue pathologies, including, e.g., the diagnosis and staging of skin and oral cancer. Combined, these efforts will improve the contrast achievable with probe-based OCT, thereby enhancing its practical use and extending its utility to new applications where decisive contrast has been lacking.