Towards Real-Time Inference of Thin Liquid Film Thickness Profiles from Interference Patterns Using Vision Transformers

TL;DR

Reconstructing thin-film thickness from interferograms is ill-posed due to $2\pi$ phase periodicity and noise. The authors propose a vision-transformer model that directly maps single interferograms to thickness profiles, trained on a hybrid dataset of synthetic and experimental tear-film data. The approach achieves real-time, single-pass inference on consumer hardware (~45 ms per frame) and is robust to motion and illumination artifacts, validated on in vivo tear-film interferograms. This work enables continuous, non-invasive tear-film monitoring and suggests applicability to other thin-film interferometry domains.

Abstract

Thin film interferometry is a powerful technique for non-invasively measuring liquid film thickness with applications in ophthalmology, but its clinical translation is hindered by the challenges in reconstructing thickness profiles from interference patterns - an ill-posed inverse problem complicated by phase periodicity, imaging noise and ambient artifacts. Traditional reconstruction methods are either computationally intensive, sensitive to noise, or require manual expert analysis, which is impractical for real-time diagnostics. To address this challenge, here we present a vision transformer-based approach for real-time inference of thin liquid film thickness profiles directly from isolated interferograms. Trained on a hybrid dataset combining physiologically-relevant synthetic and experimental tear film data, our model leverages long-range spatial correlations to resolve phase ambiguities and reconstruct temporally coherent thickness profiles in a single forward pass from dynamic interferograms acquired in vivo and ex vivo. The network demonstrates state-of-the-art performance on noisy, rapidly-evolving films with motion artifacts, overcoming limitations of conventional phase-unwrapping and iterative fitting methods. Our data-driven approach enables automated, consistent thickness reconstruction at real-time speeds on consumer hardware, opening new possibilities for continuous monitoring of pre-lens ocular tear films and non-invasive diagnosis of conditions such as the dry eye disease.

Figures (2)

• Figure 1: Experimental setup and data generation. ( a) Schematic of the Tear Film Interferometer showing dome light illumination, lens assembly, and camera. ( b) In vivo eye interferogram with reconstructed tear film thickness profile. ( c) Synthetic data generation workflow: synthetically generated thickness profiles are mapped to color interferograms using a reference colormap to generate training interferograms-thickness profile pairs. ( d) Mean thickness distribution in the training set. ( e) Thickness range distribution in the training set.
• Figure 2: Model performance on experimental and synthetic data. (a) Example in vivo tear film interferogram, predicted thickness profile, and temporal comparison between predicted and ground truth mean thickness. (b) Effect of dataset size on model accuracy, showing RMSE versus synthetic to experiment data split ratios for datasets totaling 5k, 10k, and 25k samples. (c) Influence of transformer model sizes (Tiny–Large) on prediction accuracy as a function of synthetic to experiment data split ratios for a dataset totaling 5k images.