NeuralOCT: Airway OCT Analysis via Neural Fields

TL;DR

This work tackles the challenge of obtaining high-fidelity 3D airway geometries from endoscopic OCT in pediatric populations without radiation. It introduces NeuralOCT, a two-stage pipeline that first uses 2D segmentation (teacher) to generate wall point clouds and then optimizes a coordinate-based neural field (student) to represent the surface as a signed distance function, trained with NeuralPull to align surface samples. The approach is the first to recover 3D geometries from raw segmentation in airway OCT and to employ neural fields for OCT geometry, achieving sub-0.07 mm average A-line error and producing robust, smooth reconstructions that outperform 2D baselines in downstream 3D metrics. The method advances shape analysis and enables potential surgical simulations, with code to be released on GitHub.

Abstract

Optical coherence tomography (OCT) is a popular modality in ophthalmology and is also used intravascularly. Our interest in this work is OCT in the context of airway abnormalities in infants and children where the high resolution of OCT and the fact that it is radiation-free is important. The goal of airway OCT is to provide accurate estimates of airway geometry (in 2D and 3D) to assess airway abnormalities such as subglottic stenosis. We propose $\texttt{NeuralOCT}$, a learning-based approach to process airway OCT images. Specifically, $\texttt{NeuralOCT}$ extracts 3D geometries from OCT scans by robustly bridging two steps: point cloud extraction via 2D segmentation and 3D reconstruction from point clouds via neural fields. Our experiments show that $\texttt{NeuralOCT}$ produces accurate and robust 3D airway reconstructions with an average A-line error smaller than 70 micrometer. Our code will cbe available on GitHub.

Figures (4)

• Figure 1: Principle of the aOCT scanning process. (a) describes the airway OCT scanning process, during which frames are captured as the catheter is pulled back from the bottom to the top of the airway. In (b), laser rays are emitted from the catheter and hit the airway wall to estimate line-of-sight distance as helical A-lines. (c) illustrates how to relate the coordinates on the airway wall to the catheter and laser geometry in a cylindrical coordinate system.
• Figure 2: Principle of the NeuralOCT approach. NeuralOCT extracts 3D geometries from OCT scans by combining point cloud extraction from 2D segmentations with 3D reconstruction via neural fields.
• Figure 3: Visualizations of the airway OCT segmentation. (a,c) compares the predicted light-of-sight distances from the teacher module with different methods on scan A and scan B; (b,d) compares the predicted light-of-sight distances from the student module on scan A and scan B. More visualizations are available in the supplementary material.
• Figure 4: Visualizations of the 3D airway geometry reconstructions. Each pair of shapes represents the raw point clouds from the teacher module and 3D reconstructions from the student module respectively.