Enhanced Feature Segmentation and Data Extraction in Pentacam Scheimpflug Images

Abstract

Purpose: : Scheimpflug imaging using Pentacam is widely used to assess straylight of cornea and lens. In a number of cases the system's depth penetration does not allow for a reliable imaging down to the back surface of the lens rendering the inbuilt segmentation function inapplicable. To provide automated segmentation for a wide range of imaging conditions a new method was implemented and tested. The extracted data enables the calculation of scatter related characteristics.

Methods: : 254 Pentacam images from 136 subjects (age 44±16 years) under mydriasis were included in the procedure. Exclusion criteria were artifacts like shadowing lids and lashes disturbing a symmetric lens image. For comparison 2 vertical line profiles were placed near the optical axis using the Pentacam software. Raw image data was extracted. A Matlab implementation of segmentation algorithms starts with automated detection of the innermost iris points as a prominent feature in every raw image of sufficient quality. In some cases user interaction may be necessary to locate the iris points. Thresholding based on background level yields back and front surfaces of lens and cornea. The respective boundaries are generated by polynomial fitting, outlier clipping and in case of the lens the application of a convex hull. Lateral boundaries are defined by vertical clipping relative to the iris points, assuming the influence on density measurements to be negligible. From the segmented areas intensity histograms are extracted as well as vertical and horizontal line profiles at well defined lens positions.

Results: : According to visual inspection the segmentation method worked apparently well on all images provided in the series. For objective evaluation manual line profile means were compared to automatic generated profiles yielding a good correlation of R²=0.93. Enhanced characteristics were derived from extracted data like lens histogram width and lens center intensity which correlated well with age (R²=0.76 and R²=0.8 respectively) in exponential fit. The increase with age was more pronounced for said characteristics (e^0.012 and e^0.025 respectively) than for manual profile means (e^0.008).

Conclusions: : An automated segmentation method for Pentacam Scheimpflug images enables densitometric measurements on an extended range of images. Additional scatter characteristics were calculated which are subject to further studies on applicability and reproducibility.

Keywords: imaging/image analysis: non-clinical • anterior segment • cataract