Variability Between Trained Experts Performing Hand-Drawn Segmentations of CirrusTm HD-OCT Scans

Abstract

 

Purpose:

 

To determine the differences between hand-drawn segmentations performed by three trained experts on images from Cirrus^TM HD-OCT, a spectral domain optical coherence tomography instrument (Carl Zeiss Meditec Inc., Dublin, CA).

 

Methods:

 

24 Cirrus^TM HD-OCT scans were randomly chosen from a database of 163 scans, classified by scan type and disease. Scans were done on consented subjects, following an IRB approved protocol. Hand-drawn segmentations on the internal limiting membrane (ILM) and retinal pigment epithelium (RPE) boundaries were performed by 3 trained experts on each scan. Differences between the three hand drawn segmentations were measured.

 

Results:

 

The mean absolute differences (MAD) averaged over all points in all scans are shown in Table 1. Specific cases were analyzed to understand factors causing these differences.Table 1. Mean absolute difference between hand drawn segmentations of three trained experts  

 

Conclusions:

 

In most disease categories, especially for the higher density 512x128 scans, there was good agreement between the three experts (MAD < 6 microns). Greater variability seen in specific groups appear to be influenced by various factors, such as individual interpretations of retinal anatomy, changes due to disease, scan type, image quality and amount of training. This variability should be taken into consideration as a contributing factor to differences seen when using hand drawn segmentations to validate the instrument’s automated segmentation algorithm.

 

Keywords: imaging/image analysis: clinical • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • retina