Harikrishnan Unnikrishnan

arXiv2026

Optimal Sampling and Noise Mitigation for Vocal Fold Kinematics in High-Speed Videoendoscopy

Objectives: High-Speed Videoendoscopy glottal area waveforms conflate lateral fold excursion with longitudinal opening, precluding direct estimation of velocity, acceleration, and left-right asymmetry. Displacement waveforms preserve these quantities but are susceptible to spatial quantization and temporal sampling errors. We characterize these errors, evaluate reduction via Kalman smoothing and temporal interpolation, and determine optimal frame rate bounds relative to pixel excursion, speed quotient (SQ), and open quotient (OQ). Methods: Sampling theory was used to derive analytical frame-rate bounds required for kinematic feature fidelity. Kalman smoothing was compared with FFT-based denoising, paired with sinc or spline interpolation. Interpolation was evaluated on 4000 fps recordings (N=55) downsampled and reconstructed. A 16000 fps dataset (N=12) served as reference for noise reduction across 2000-8000 fps downsampled waveforms evaluated on peak displacement, velocity, OQ, and SQ. Results: Sinc interpolation yielded lower median mean absolute error than spline (0.29 vs 0.52 pixels), with 85% of recordings (47/55) versus 45% (25/55) below 0.5 pixels. For kinematic feature estimation (N=12), Kalman-sinc outperformed FFT-sinc in 75% of feature-by-frame-rate pairings. Conclusion: Computing kinematic features without artifacts requires joint consideration of temporal and spatial resolution. Kalman-sinc preprocessing at 8000 fps is recommended for velocity and SQ measures. Peak lateral displacement can be measured at 2000 fps, while OQ requires 4000 fps for typical spatial resolutions (9-14 pixels maximum excursion). Pixel excursion should be reported alongside kinematic measures to contextualize data reliability.

eess.IV↗

arXiv2026

A Detection-Gated Pipeline for Robust Glottal Area Waveform Extraction and Clinical Pathology Assessment

We present a fully automated, two-stage modular glottal area segmentation framework for high-speed videoendoscopy (HSV) designed for accuracy, generalizability, and real-time playback. Our detection-gated pipeline combines a YOLOv8n glottis localizer with a U-Net segmenter; the localizer defines a tight crop to ensure a consistent field of view and gates the output to reduce spurious segmentations during glottal closure. The models were trained on the GIRAFE (N=600) and BAGLS (N=55,750) datasets. Cross-dataset portability was evaluated by benchmarking GIRAFE-trained models on the BAGLS test set without fine-tuning. In these evaluations, the pipeline achieved a Dice Similarity Coefficient (DSC) of 0.745 (87% of the in-domain ceiling). On in-distribution test sets, the system achieved DSCs of 0.81 (GIRAFE) and 0.856 (BAGLS), outperforming or competing with state-of-the-art methods. An exploratory clinical study of 40 subjects demonstrated that the glottal area Coefficient of Variation (CV distinguished healthy from pathological function (p=0.006). The system processes ~35 frames per second on commodity hardware, enabling interactive clinical review. This design supports uniform extraction of laryngeal kinematic measures across varying acquisition settings. Code, weights, and software are available at https://github.com/hari-krishnan/openglottal.

cs.CV↗