Automated detection of Lamina Cribrosa (LC) beam microarchitecture from imaging data using a Frangi filter

Abstract

 

Purpose

 

To develop a robust, automated detection method for quantification of LC beam microarchitecture from imaging data, such as micro CT or multiphoton microscopy. Biomechanically-mediated insult to cells within the LC is thought to play an important role in the pathogenesis of glaucoma, implying the need for improved understanding of the structural and micro-mechanical environment in the LC.

 

Methods

 

The Frangi filter image processing technique, originally proposed for automated detection of vessels in angiographic images (MICCAI’98, 1496:130, 1998), identifies the eigenvalues of the Hessian (essentially the second derivative) of voxel intensity. It returns the probability that a given voxel lies within a “blob-like”, “tube-like”, or “plate-like” structure. Post mortem porcine and human eyes were fixed in 4% PFA and incubated with phosphotungstic acid (PTA) for 21 days to enhance contrast. Their posterior segments were imaged using micro-computed tomography (μCT) with 5 μm isotropic spatial resolution (180 keV, 185 μA, 6284 projections). LC boundaries were manually delimited in 3D data sets, and the Frangi filter was applied; voxels with a “beam probability” of > 0.2 were empirically determined to lie in beams.

 

Results

 

The Frangi filter successfully identified the plate-like structures comprising the porcine LC (see figure). It is important to note that this approach uses a fully 3D data set, runs in order of minutes, and provides information on local beam orientations (figure inset). We describe elsewhere how this information can be used to drive biomechanical models of the LC.

 

Conclusions

 

The Frangi filter holds significant promise for fast quantification of LC microarchitecture from imaging data. Specifically, when combined with ongoing techniques for imaging the LC under native conditions at various IOP levels, the Frangi filter will provide essential information about tissue deformation and biomechanical environment under pressure elevation.

 

 

LC beam segmentation using Frangi filter. Left panel shows single slice through 3D μCT image of porcine LC. Upper images show “beam probability” in greyscale (left) and segmented image (right). Lower images show beam orientation information overview (left) and zoom (right). Although a single slice is shown, all operations are carried out in 3D across the entire LC.

 

Keywords: 577 lamina cribrosa • 552 imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • 551 imaging/image analysis: non-clinical