Real-time corneal segmentation and 3D needle tracking in intrasurgical OCT

Abstract

Ophthalmic procedures demand precise surgical instrument control in depth, yet standard operating microscopes supply limited depth perception. Current commercial microscope-integrated optical coherence tomography partially meets this need with manually-positioned cross-sectional images that offer qualitative estimates of depth. In this work, we present methods for automatic quantitative depth measurement using real-time, two-surface corneal segmentation and needle tracking in OCT volumes. We then demonstrate these methods for guidance of ex vivo deep anterior lamellar keratoplasty (DALK) needle insertions. Surgeons using the output of these methods improved their ability to reach a target depth, and decreased their incidence of corneal perforations, both with statistical significance. We believe these methods could increase the success rate of DALK and thereby improve patient outcomes.

Keywords: (100.0100) Image processing; (110.4500) Optical coherence tomography; (170.1610) Clinical applications.

Fig. 1

Flow chart of segmentation and tracking to find needle penetration depth. The acquisition software acquired volumes of 96 B-scans in three groups of 32 B-scans. B-scan segmentation of each group occurred during the acquisition of the next group.

Fig. 2

Illustration of corneal segmentation. (A) Original image obtained from human cadaver corneal sample. (B) Epithelial segmentation (orange) with epithelial constraints (magenta). (C) Epithelial and endothelial segmentation (orange) with endothelial constraints (magenta).

Fig. 3

Representation of the process used to determine an estimate of the needle base and tip. (A) DC-subtracted maximum intensity projection (MIP). (B) Thresholded depth map. (C) Six largest connected components from the depth map. Only the green connected component fit the width criterion. (D) Needle base estimate (green circle) and needle tip estimate (red circle) based on the intersection of the line formed by the first principal component with the borders of the image (blue line). Pixels identified as the needle are orange. Best viewed in color.

Fig. 4

Example needle shadow segmentation correction. (A) B-scan with uncorrected segmentation. Shadows from the needle interfere with the endothelial surface segmentation. Inflating the needle allows for the area by the white arrow to be corrected. (B) Corrected segmentation taken from height map in (F). (C) Height map of the endothelial surface of the original segmentation. Black arrow denotes corrupted segmentation caused by the needle. (D) Height map of the endothelial surface of the original segmentation with the inflated needle pixels marked in green and the location of B-scan (A) and (B) denoted by the blue line. (E) Height map of the endothelial surface of the original segmentation with pixels that changed after the trial inpainting marked in green. (F) Corrected height map after inpainting green pixels in (E). Black arrow denotes original location of corrupted segmentation.

Fig. 5

Refraction corrected cross section along the axis of the needle. Green dots denote the epithelial surface point, needle tip, and endothelial surface point used to compute the depth along the magenta line.

Fig. 6

Experimental setup for validation experiments. In the experiment where corneal fellows inserted needles into the cornea, a tracked cross section was displayed on the monitor next to the microscope.

Fig. 7

Series of images depicting different needle penetration depths, as shown to all surgeons prior to performing the experiment. Needle percent depths are displayed at the bottom of each image.

Fig. 8

Comparison of manual and automatic segmentation for a B-scan with and without a needle. (A) Original B-scan, with no needle. (B) Segmented B-scan. Green denotes the manual segmentation and purple denotes the automatic segmentation. Where the green is not visible, the two methods segmented the same point. (C) Original B-scan with a needle. (D) Uncorrected automatic segmentation. (E) Corrected automatic segmentation (purple) and manual segmentation (green). Best viewed in color.

Fig. 9

(A) Plot of the final needle depth expressed as a percent of corneal thickness for all trials in which the surgeon did not puncture the endothelium. A blue X indicates the mean of the group and error bars denote one standard deviation. (B) Plot illustrating performance of the automatic needle percent depth calculation compared to the manual calculation.