Abstract
Purpose :
This abstract describes an efficient method to evaluate the level of contrast in data obtained from OCT systems, in which the quality and usefulness of the data are determined prior to any algorithmic analyses such as segmentation. The metric derived from the algorithm is a measure of feature strength in the OCT dataset which can be distinguished from the signal strength or SNR. For example, a blurry image can have high signal strength with low quality due to poor contrast.
Methods :
The integral of the absolute gradients in the axial direction for a given A-scan (or a portion of it) indicates the overall contrast of that A-scan. The gradients in the axial direction are computed by convolving each A-scan data with the derivative of a Gaussian function. A nonlinear transform such as generalized logistic function scales the gradients data and eliminates noise. An integral map from the volumetric OCT dataset is generated by integrating the absolute gradients in the axial direction at each lateral scan position. The integral map is converted to a probability map by transforming the map data using a cumulative distribution function (CDF) determined from integral maps of around 5000 volumetric OCT data with variety of scan qualities acquired using CIRRUS™ HD-OCT 5000 with AngioPlex® OCT Angiography (ZEISS, Dublin, CA). Average value of a probability map in relevant scan regions indicates the quality score ([0 1]) of an OCT dataset (Fig 1).
Results :
Fig 2 shows examples of probability maps and segmentation failure or inaccuracy in the regions with low quality. The examples show that the performance of the segmentation of a specific layer boundary depends on the level of contrast in the local regions of OCT data.
Conclusions :
An efficient method of displaying and evaluating the level of contrast of an OCT dataset is presented. The method creates a probability map that shows partitions of a volumetric OCT dataset with good and poor qualities, which helps to reduce spurious subsequent analyses of the data.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.