Comparison of manual and automated 3D measurements of ciliary body in three dimensional ultrasound biomicroscopy (3D-UBM) images.

Abstract

Purpose : We developed 3D-UBM imaging technique to assess the 3D morphology of ciliary body which plays crucial role in glaucoma.

Methods : We acquired 3D-UBM volumes from cadaver eyes and applied slice alignment based on normalized cross-correlation and image denoising via frame averaging. Volume was aligned perpendicular to the optic axis. From the aligned volumes, en face images were extracted and ciliary muscle in those images were manually annotated by 2 readers and verified by an expert. Training data include en face images from 5 cadaver eye volumes and the ground truth from the expert. U-Net was trained for automatic segmentation of ciliary muscle. From the ground truth and predicted segmentation, we can make volumetric measurements of ciliary body. Predicted segmentations on new volumes are being sent to readers for editing, so as to generate a larger dataset for training, creating an active learning paradigm.

Results : Automatic segmentation of ciliary muscle in en face images gave a Dice score of 0.72 ± 0.08. The model predicts thicker boundary in en face images where the ciliary muscle is very narrow and predicts newly identified ciliary muscle region, especially deeper into the tissue which are not missed as ground truth due to degraded image contrast. From the manual segmentation of an ex vivo eye, we measured ciliary body volume of 67.87 mm³, muscle volume of 52.73 mm³, average ciliary process surface of 3.02mm². These measurements cannot be made from conventional 2D UBM system. From the manual and automated segmentation, we measure the volume of ciliary muscle and area of ciliary muscle in en face images. We see an average error of ~19.25%, which can be attributed to thicker boundary and newly identified regions. In the en-face area measurements between automated and manual method, we see an average intra-class correlation coefficient of 0.86, showing good agreeability in area measurements.

Conclusions : Three-dimensional segmentation of ciliary muscle is enabling unique assessments of biometrics which are more accurate than can be obtained using conventional 2D UBM. Such 3D assessments will reveal effects of glaucoma medication and surgery.

This is a 2020 Imaging in the Eye Conference abstract.

 

Figure 1. Automated segmentation of ciliary muscle using 3D-UBM images.

 

Figure 2. 3D rendering of the ciliary body from which we can compute 3D measurements i.e. surface area, enclosed volume.