Abstract
Purpose :
This study investigates an artificial intelligence (AI) assisted method to automatically identify and classify vitreous cells in uveitis with ultrahigh-resolution (UHR) optical coherence tomography (OCT). The immune system plays a major role in the inflammatory response; having a non-invasive tool to quantify different types of vitreous cells can be a game changer in uveitis care.
Methods :
A 250-kHz retinal UHR-OCT prototype with 1.8 µm axial resolution (full-width-half-maximum in tissue) was used to image cells in vitro and in vivo. Leukocytes were separated from a human blood sample and sorted with a flow cytometer. Cell suspensions of neutrophils, lymphocytes, and monocytes were placed into cuvettes and imaged with OCT. A volumetric scan pattern (1 mm × 1 mm, 500 raster lines of 1000 axial scans each) was used. Custom-designed software algorithms were developed to automatically locate cells in the OCT volume as hyperreflective spots. For each cell, a 3-dimensional volume of 16×8×4 (axial × transverse × lateral) voxels, whose center matched the center of individual cells, were extracted from the OCT image. A lightweight convolutional neural network (CNN) model was developed to classify the types of inflammatory cells (details in a separate ARVO abstract). In the clinical study, the posterior vitreous of patients with panuveitis or posterior uveitis and active inflammation was imaged with OCT. Vitreous cells were detected and analyzed by the same automated software. The percentage composition of the vitreous inflammatory cells was estimated.
Results :
Our study scanned three patients with posterior vitreous cells. One patient was diagnosed with Birdshot chorioretinopathy, one with chronic posterior uveitis, and the other one with chronic panuveitis. The CNN model detected predominantly lymphocytes (range 60.5% to 94.4%) in UHR-OCT images of the posterior vitreous for all three patients. These results aligned with the clinical information.
Conclusions :
AI-assisted UHR-OCT objectively quantified and classified posterior inflammatory cells. It can provide a useful non-invasive diagnostic biomarker to aid uveitis treatment decisions, understand the clinical course of diseases, and follow patients over time. Future studies are needed to investigate this method with a larger patient sample size and more uveitis disease diversity.
This abstract was presented at the 2024 ARVO Imaging in the Eye Conference, held in Seattle, WA, May 4, 2024.