Automated Detection of IS/OS Defect Regions in 3D OCT Images

Abstract

 

Purpose

 

Defect of Inner Segment/Outer Segment (IS/OS) on SD-OCT is associated with lower visual acuity outcome in patients suffering from trauma, probably related to the destruction of the rod and cone cells in the photoreceptor. However, no automated method to estimate IS/OS defect from SD-OCT exists.

 

Methods

 

9 subjects diagnosed with trauma were included and underwent macular-centered SD-OCT (Topcon, 512×64×480 voxels, 11.72×93.75×3.50µm3, or 512×128×480 voxels, 11.72×46.88×3.50µm3). Automated detection of IS/OS defect was realized as follows. First, eleven surfaces were automatically segmented using the multi-scaled 3D graph search approach [1], the sub-volume between surface 7 and 8 containing the IS/OS region was flattened based on the segmented retinal pigment epithelium (RPE) layer. If the IS/OS is disrupted around the central fovea, it will greatly implicate the patient’s central visual acuity. In this preliminary research, the sub-volumes centered at the fovea (diameter = 1mm) between surface 7 and surface 8 are our volumes of interest (VOIs).Second, each voxel in the VOIs was classified into disrupted or not disrupted based on 5 types of textural features, using a KNN classifier. The ground truth was manually marked by two observers independently, under the supervision of an experienced ophthalmologist. The performance of the proposed method were analyzed.

 

Results

 

In patients with trauma, large defect regions of the IS/OS can be detected using the proposed method (Fig. 1). The detection results are accordance with the ground truth well, the performance of true positive rate and true negative rate is 0.802 and 0.796, respectively.

 

Conclusions

 

Automated detection of IS/OS defect in patients with trauma has been achieved. Though this pilot study concerned Topcon scans, the algorithm can be applied to SD-OCT images from any manufacturer. We have started determining the relationship of quantitative IS/OS defect to visual acuity in this type of patient.

 

 

Figure.1 An example of IS/OS defect region detection. (1) One slice of the flattened OCT image; (2) Segmentation results of 11 intra-retinal surfaces; (3) VOIs extraction; (4) the detection results of the IS/OS defect region (only 5 continuous slices of VOIs are shown). White voxel represents the disrupted region. The upper are the ground truth and the lower are the detection results of the proposed method.

 

Keywords: 550 imaging/image analysis: clinical • 688 retina