Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Real-time topology-aware M-mode OCT segmentation for robotic deep anterior lamellar keratoplasty (DALK) guidance

Rosalinda Xiong, Jinglun Yu, Yaning Wang, Ziyi Huang, Jin U. Kang

TL;DR

This work tackles the challenge of providing real-time, depth-aware guidance for robotic DALK by delivering topology-aware M-mode OCT segmentation that remains robust under speckle noise and instrument shadowing. It introduces a topology-regularized UNeXt framework that processes M-mode frames in stripes and optimizes a combined BCE, Dice, and star-shaped topology loss to enforce anatomically consistent boundaries, all within an end-to-end pipeline capable of exceeding $80~\mathrm{Hz}$ throughput. Evaluation on a rabbit-eye M-mode OCT dataset shows improved boundary stability relative to topology-agnostic baselines, with maintained deployable frame rates and resilience to transient signal degradation. The method offers practical real-time depth overlays for surgical guidance and suggests future enhancements for confidence gating and broader validation across systems.

Abstract

Robotic deep anterior lamellar keratoplasty (DALK) requires accurate real time depth feedback to approach Descemet's membrane (DM) without perforation. M-mode intraoperative optical coherence tomography (OCT) provides high temporal resolution depth traces, but speckle noise, attenuation, and instrument induced shadowing often result in discontinuous or ambiguous layer interfaces that challenge anatomically consistent segmentation at deployment frame rates. We present a lightweight, topology aware M-mode segmentation pipeline based on UNeXt that incorporates anatomical topology regularization to stabilize boundary continuity and layer ordering under low signal to noise ratio conditions. The proposed system achieves end to end throughput exceeding 80 Hz measured over the complete preprocessing inference overlay pipeline on a single GPU, demonstrating practical real time guidance beyond model only timing. This operating margin provides temporal headroom to reject low quality or dropout frames while maintaining a stable effective depth update rate. Evaluation on a standard rabbit eye M-mode dataset using an established baseline protocol shows improved qualitative boundary stability compared with topology agnostic controls, while preserving deployable real time performance.

Real-time topology-aware M-mode OCT segmentation for robotic deep anterior lamellar keratoplasty (DALK) guidance

TL;DR

This work tackles the challenge of providing real-time, depth-aware guidance for robotic DALK by delivering topology-aware M-mode OCT segmentation that remains robust under speckle noise and instrument shadowing. It introduces a topology-regularized UNeXt framework that processes M-mode frames in stripes and optimizes a combined BCE, Dice, and star-shaped topology loss to enforce anatomically consistent boundaries, all within an end-to-end pipeline capable of exceeding throughput. Evaluation on a rabbit-eye M-mode OCT dataset shows improved boundary stability relative to topology-agnostic baselines, with maintained deployable frame rates and resilience to transient signal degradation. The method offers practical real-time depth overlays for surgical guidance and suggests future enhancements for confidence gating and broader validation across systems.

Abstract

Robotic deep anterior lamellar keratoplasty (DALK) requires accurate real time depth feedback to approach Descemet's membrane (DM) without perforation. M-mode intraoperative optical coherence tomography (OCT) provides high temporal resolution depth traces, but speckle noise, attenuation, and instrument induced shadowing often result in discontinuous or ambiguous layer interfaces that challenge anatomically consistent segmentation at deployment frame rates. We present a lightweight, topology aware M-mode segmentation pipeline based on UNeXt that incorporates anatomical topology regularization to stabilize boundary continuity and layer ordering under low signal to noise ratio conditions. The proposed system achieves end to end throughput exceeding 80 Hz measured over the complete preprocessing inference overlay pipeline on a single GPU, demonstrating practical real time guidance beyond model only timing. This operating margin provides temporal headroom to reject low quality or dropout frames while maintaining a stable effective depth update rate. Evaluation on a standard rabbit eye M-mode dataset using an established baseline protocol shows improved qualitative boundary stability compared with topology agnostic controls, while preserving deployable real time performance.
Paper Structure (12 sections, 2 equations, 1 figure, 2 tables)

This paper contains 12 sections, 2 equations, 1 figure, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. METHODS
  3. Dataset and Preprocessing
  4. Model Architecture and Inference
  5. End-to-End Deployment Pipeline
  6. Loss Function
  7. Evaluation Metrics
  8. EXPERIMENTS AND RESULTS
  9. Implementation Details
  10. Prior Method Reference Performance
  11. Temporal Robustness and Dropout Tolerance
  12. DISCUSSION AND CONCLUSION

Figures (1)

  • Figure 1: Representative solid-band overlays produced by the proposed topology-aware UNeXt pipeline. Topology regularization reduces boundary discontinuities and suppresses anatomically implausible layer crossings under low-SNR conditions.