Three-dimensional visualization of intrachoroidal cavitation using the deep learning-based enhancement of optical coherence tomography

Abstract

Purpose : The intrachoroidal cavitation (ICC) is a peripapillary pathological lesion usually associated with high myopia, which may cause serious clinical consequences such as visual field defect. However, due to the challenges in existing imaging devices to visualize three-dimensional (3D) morphology of the ICC, details in structural characteristics and their influence on visual function, remain poorly understood. Using deep learning (DL)-based image enhancement and 3D rendering of volumetric swept-source optical coherence tomography (SSOCT) images, we evaluated the 3D anatomic characteristics of the ICC in comparison with the existing 2D parameters.

Methods : Thirteen eyes of 12 consecutive patients with peripapillary ICC were enrolled (age: 52.5±6.3 years, refractive error: -8.4±3.2 D, and axial length: 27.1±1.4 mm). DL-based image enhancement, automatic segmentation of the ICC, and 3D rendering was applied to parapapillary 6 x 6 mm volumetric SSOCT scans. We evaluated the anatomical relationship between the ICC and the Bruch membrane opening (BMO). The 3D volume, as well as the 2D maximum width and length of the ICC, were calculated as morphologic parameters. Associations between the ICC parameters and visual field mean deviation (VFMD) of standard automated perimetry were investigated.

Results : The ICC was successfully segmented, and 3D visualized in all eyes. The ICC was located inferior to the optic disc in all eyes and observed bilaterally in 1 patient and unilaterally in 11 patients. The ICC was detected with fundus photographs in 12 eyes out of the 13 eyes. VFMD ranged from -11.1 to 0.12 dB (mean -2.61±3.16 dB). All the ICCs showed direct contact with the BMOs and two cases showed the overlap. The ICC volume ranged from 55.3 to 975.4 x10^-3 mm³ (358.4±283.3 x10^-3 mm³), the ICC width from 118.5 to 512.4 µm (248.5±108.9 µm), and the ICC length from 0.19 to 2.00 mm (1.10±0.47 mm). The ICC volume (P=0.02, regression coefficient=-0.007) was significantly associated with VFMD while the ICC width and length were not.

Conclusions : 3D images of the ICC were successfully generated using DL-based enhancement of SSOCT volumetric scans. 3D volume parameter seems to reflect the pathological influence of the ICC on visual function better than the existing 2D metrics, which potentially leads to an improved pathological understanding of the ICC.

This is a 2021 ARVO Annual Meeting abstract.