Dynamic OCT tracking for enhanced visualization of ophthalmic surgery

About This Grant

PROJECT ABSTRACT Cataracts and glaucoma are the two leading causes of blindness worldwide. Crucial ophthalmic procedures to treat cataract, glaucoma, and other vision conditions require precise visualization of anatomy and microsurgical instruments. Visualization in such surgeries has been limited to stereo optical microscopes since the early 20th century. With advancements in optical coherence tomography (OCT), we can now obtain real-time 3D visualization within the eye. Over the past decade, intraoperative OCT (iOCT) systems have become widely researched and integrated into the latest ophthalmic microscopes built by companies such as Zeiss and Leica. These iOCT systems come with the potential to revolutionize ophthalmic surgery, with an unparalleled ability to resolve key anatomic features at micron-level precision. However, there is a crucial challenge that hampers the clinical utility of iOCT. This challenge stems from the fundamental tradeoff between OCT field-of-view and imaging speed. This tradeoff constrains state-of-the-art systems to operate with a relatively small (e.g. 5x5 mm) field of view to achieve the volume update speeds (~10-15 Hz) required for surgical visualization. Consequently, a trained operator on the surgical team must manually reposition the OCT scan throughout the surgery. The current implementation of iOCT results in a “point-and-shoot” approach to imaging, i.e. using OCT as an intermittent snapshot tool, rather than as a continuous surgical visualization technology. With even small movements of the surgical instruments, the OCT image can quickly lose sight of the surgical region of interest (ROI). Manual tracking of iOCT discards a key advantage of OCT, which is real-time 3D data collection. With advances in deep learning methods for image processing and object recognition, there are new opportunities to tackle this problem. The goal of this project is to engineer a novel computational system for automatic, real- time tracking of the surgical ROI in a clinical iOCT system. Our vision is to develop a system that can be readily applied to existing clinical microscopes, and adaptable to future robotic surgical systems. As part of our preliminary work, we have created a lateral tool tracking OCT system using deep learning models applied to the microscope feed. Our current system utilizes a novel synthetic data approach, making use of 3D-rendered models of eyes and tools to accelerate deep learning model development. In the proposed project, we expand on this preliminary work by developing a system for 3D multimodal surgical ROI tracking of iOCT that can be applied to many different types of ophthalmic surgeries. We will then evaluate our platform via ex-vivo porcine and human cadaver eye studies with wet-lab benchmarking and simulated surgeries with our clinical collaborators. Our immediate application is ophthalmic surgery, but the methodology has relevance to a wide range of 3D imaging systems for microsurgical procedures. By developing this system for dynamic OCT surgical tracking, we hope to improve ophthalmic visualization in both training and surgical practice.