The Use of Artificial Intelligence for Estimating Anterior Chamber Depth from Slit-Lamp Images Developed Using Anterior-Segment Optical Coherence Tomography

Abstract

Primary angle closure glaucoma (PACG) is a major cause of visual impairment, particularly in Asia. Although effective screening tools are necessary, the current gold standard is complex and time-consuming, requiring extensive expertise. Artificial intelligence has introduced new opportunities for innovation in ophthalmic imaging. Anterior chamber depth (ACD) is a key risk factor for angle closure and has been suggested as a quick screening parameter for PACG. This study aims to develop an AI algorithm to quantitatively predict ACD from anterior segment photographs captured using a portable smartphone slit-lamp microscope. We retrospectively collected 204,639 frames from 1586 eyes, with ACD values obtained by anterior-segment OCT. We developed two models, (Model 1) diagnosable frame extraction and (Model 2) ACD estimation, using SWSL ResNet as the machine learning model. Model 1 achieved an accuracy of 0.994. Model 2 achieved an MAE of 0.093 ± 0.082 mm, an MSE of 0.123 ± 0.170 mm, and a correlation of R = 0.953. Furthermore, our model's estimation of the risk for angle closure showed a sensitivity of 0.943, specificity of 0.902, and an area under the curve (AUC) of 0.923 (95%CI: 0.878-0.968). We successfully developed a high-performance ACD estimation model, laying the groundwork for predicting other quantitative measurements relevant to PACG screening.

Keywords: Smart Eye Camera; algorithm; anterior chamber depth; anterior-segment optical coherence tomography; artificial intelligence; deep learning; glaucoma; machine learning; slit-lamp images; telemedicine.

Figure 1

Study design and model developments. (A) Dataset and preprocessing of study chart. (B) Diagnosable frame extraction (Model 1). Images classified as diagnosable and non-diagnosable. (C) Anterior chamber depth annotation and machine learning (Model 2). Dataset is split into training, validation, and test datasets. (D) Architecture of our machine learning models.

Figure 2

Distribution of anterior chamber depth (ACD) in our study. The x-axis represents the ACD in millimeters (mm), ranging from 1.800 mm to 4.000 mm, divided into 0.100 mm increments. The y-axis indicates the number of eyes corresponding to each ACD measurement range. The average ACD across the study population is 3.018 ± 0.385 mm.

Figure 3

Performance metrics for diagnosable frame extraction (Model 1) and anterior chamber depth (ACD) estimation (Model 2). Figure 3 presents the performance metrics for the two models used in the study. Model 1 focuses on extracting diagnosable frames. Model 2 evaluates the accuracy of ACD estimation.

Figure 4

Correlation between AS-OCT anterior chamber depth (ACD) values and AI-estimated ACD measurements. Figure 4 illustrates the correlation between AS-OCT-measured ACD and the AI-estimated ACD from both individual frames and eyes.

Figure 5

Diagnostic performance of anterior chamber depth (ACD) cut-off values for angle closure glaucoma. Figure 5 highlights the comparison of diagnostic performance metrics for different ACD cut-off values in terms of sensitivity, specificity, accuracy, and area under the curve (AUC).

Figure 6

Visualizations of anterior chamber depth (ACD) estimation model. These visualizations provide insights into the areas of the eye that are most influential in the model’s ACD estimation process using gradient-weighted class activation mapping (Grad-CAM) method. It shows that the heatmap images pointing out where the anterior chamber is.