Fast fiber shape sensing with high accuracy by a two-layer LSTM network.

The visualization and navigation of minimally invasive instruments within confined internal spaces are critical to the success of the surgery, but traditional navigation methods face challenges from electromagnetic interference and ionizing radiation. To address these limitations, we propose a fast fiber shape sensing approach with high accuracy based on a two-layer long short-term memory (LSTM) network. The model directly regresses the 3D coordinates along the fiber sensor from the Bragg wavelength shifts of fiber Bragg gratings (FBGs), bypassing explicit curvature estimation and numerical integration. In dynamic experiments, the method achieves real-time inference at 32 fps with an end-to-end latency of approximately 7.3 ms. The mean tip error is around 2.9 mm, accounting for 0.72% of the 400 mm sensing length. The results indicate that the proposed two-layer LSTM-based method enables high-speed, high-accuracy fiber shape reconstruction, offering potential for real-time fiber navigation applications.