Real-Time 3D Needle Guidance Using Dynamic Spiral OCT Scanning and Shape-Driven Pose Extraction in Retinal Interventions

Abstract

Reliable microneedle guidance during subretinal injection remains a major clinical challenge. One of the reasons is that surgeons must infer the 3D position of the needle from a stereomicroscope that offers limited depth perception and from narrow B-scan iOCT cross-sections that cover only a fraction of the operative field. While spiral OCT scanning has been previously introduced for fast and continuous volumetric visualization of the surgical field, our work builds upon this technique by focusing on the development of a lightweight, geometry-based pose extraction framework. The proposed pipeline estimates the pose of the needle directly from the sparse 3-D point cloud generated by a single spiral OCT sweep containing just 2 000 A-scans (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\approx$</tex-math></inline-formula> 20 ms acquisition at 100 kHz). This reduction in A-scans shortens OCT acquisition time for needle tracking and, together with the elimination of pixel-wise segmentation and neural-network inference, slashes computational load and facilitates real-time deployment. A multi-pass RANSAC line fit exploits prior knowledge of needle geometry and trocar-constrained approach vectors to converge rapidly and reject outliers. Validation on <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ex vivo</i> porcine eyes demonstrates a median orientation error below <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">${3.0}^{\circ }$</tex-math></inline-formula> and a lateral tip error under 20 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\mu$</tex-math></inline-formula>m, with an end-to-end latency of 46 ms that supports pose updates beyond 20 Hz. Because each spiral volume produces a local point cloud, successive sweeps can be fused on-the-fly to yield a dense, continuously updated 3-D rendering of the needle and neighbouring retina in broader field. This live volumetric feedback lets surgeons see both instrument trajectory and local retinal topography at a glance, promoting safer needle advancement, more precise bleb placement, and earlier recognition of inadvertent tissue deformation than is possible with conventional 2-D en-face displays.