Fast and memory-efficient Just-Enough Interaction for retinal layer segmentation in OCT in layer-disrupting pathology

Abstract

Purpose : Previously, we had reported the minimally-interactive approach, Just-Enough Interaction (JEI), to manually correct the regional failures caused by fully automated retinal layer segmentation methods of the OCT scans in appearance-modifying retinal diseases. Since it is computationally demanding to simultaneously segment multiple surfaces from a full resolution OCT volume, we present a fast and memory-efficient multi-resolution JEI approach.

Methods : Eleven retinal layers defined by 12 surfaces are automatically segmented using the multi-resolution LOGISMOS method whose key idea is to reduce the size of a surface-searching volume using the segmentation result on each lower-resolution level for the segmentation on the next higher-resolution level. The JEI approach is performed by modifying the graph costs of a user-defined nudge line more attractive to fix regional segmentation failures in 3D (Fig. 1). Additionally, it provides an option to select the number of adjacent slices where the same modified costs are used, which increases analysis efficiency.

Results : Twenty (20) 3D macular OCT scans (200 × 200 × 1024 voxels, 6 × 6 × 2 mm³) from Retinal Vein Occlusion (RVO) subjects were acquired using a Cirrus^TM HD-OCT machine (Carl Zeiss Meditec, Inc., Dublin, CA), and their manual layer tracings were obtained by a retinal specialist. The comparison of the full resolution and multi-resolution approaches is shown in Table 1.

Conclusions : The multi-resolution approach required less memory, showed faster running speed, and required less user interaction than our previous full resolution approach. The method is another step towards fast and reliable automated analyses of OCT imaging in retinal disease. Software is available for clinical research at http://www.iibi.uiowa.edu/content/shared-software-download.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.

 

Figure 1. Example of the 3D multi-resolution JEI approach.

 

Table 1. Comparison of the full resolution and multi-resolution JEI approaches.