Automated tear film break-up time measurement for dry eye diagnosis using deep learning

Abstract

In the realm of ophthalmology, precise measurement of tear film break-up time (TBUT) plays a crucial role in diagnosing dry eye disease (DED). This study aims to introduce an automated approach utilizing artificial intelligence (AI) to mitigate subjectivity and enhance the reliability of TBUT measurement. We employed a dataset of 47 slit lamp videos for development, while a test dataset of 20 slit lamp videos was used for evaluating the proposed approach. The multistep approach for TBUT estimation involves the utilization of a Dual-Task Siamese Network for classifying video frames into tear film breakup or non-breakup categories. Subsequently, a postprocessing step incorporates a Gaussian filter to smooth the instant breakup/non-breakup predictions effectively. Applying a threshold to the smoothed predictions identifies the initiation of tear film breakup. Our proposed method demonstrates on the evaluation dataset a precise breakup/non-breakup classification of video frames, achieving an Area Under the Curve of 0.870. At the video level, we observed a strong Pearson correlation coefficient (r) of 0.81 between TBUT assessments conducted using our approach and the ground truth. These findings underscore the potential of AI-based approaches in quantifying TBUT, presenting a promising avenue for advancing diagnostic methodologies in ophthalmology.

Keywords: Artificial intelligence; Deep learning; Dry eye disease; Dual task learning; Siamese network; Tear film breakup time.

Figure 1

Sample of TBUT video frames from the four different classes.

Figure 2

The proposed Dual Task Siamese Network (DTSN) architecture for frames classification: It consists of two identical subnetworks; each pair of frames, $(x_{i,1},x_{i,2})$, is passed through these subnetworks, resulting in two feature vectors, $z_{i,1}$ and $z_{i,2}$, respectively. The contrastive loss ${\mathcal{L}}_c$ is then computed based on the Euclidean distance between $z_{i,1}$ and $z_{i,2}$, indicating the similarity between $x_{i,1}$ and $x_{i,2}$. ${\mathcal{L}}_b$ is a binary cross entropy (where $b\in \{1,2\}$ is the branch index). In the validation phase, only one branch of the networks was used as a non-breakup/breakup classifier.

Figure 3

Pipeline for TBUT estimation after frames classification. Green boxes denote actions applied at the frame level, while blue boxes signify actions applied to sequences.

Figure 4

Correlation between manually annotated TBUT and estimated TBUT with AI.

Figure 5

Comparison of Tear Film Dynamics for a given patient: Ground Truth and Predictions.

Figure 6

The error density for AI and clinical predictions versus ground truth.