Abstract
Purpose :
Clinical ultrasound biomicroscopy (UBM) systems use single element ultrasound transducer which lack dynamic focusing. Therefore, image quality outside the focal region is degraded due to point-spread function blurring. Deconvolution is an image processing approach that can enhance images by reducing the effect of such blurring. In this work, we evaluated the effect of deconvolution in biometric measurements made on UBM images.
Methods :
We collected 85 radial UBM images from 18 patients using our 3D-UBM system. Images were enhanced via a trained deep neural network-based deconvolution approach. Trabecular-iris angle (TIA), angle opening distance (AOD) and angle recess area (ARA) from pre-enhanced and post-enhanced images were measured as biometrics by two readers.
Results :
Deconvolution improves intra-reader biometric measurements. Both readers show significant difference (p<0.0001) in mean AOD and ARA measurements before and after deconvolution, but not in TIA measurements. Deconvolution may also improve inter-reader agreeability in certain biometric measurements. We observed higher agreeability in ARA measurements between readers after deconvolution. This might be due to the deconvolution-related contrast enhancement of the angle region.
Conclusions :
Deconvolution may be used as a post-processing step in ultrasound systems with no dynamic focusing.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.