Related Experiment Videos
An Automated Reference Frame Selection (ARFS) Algorithm for Cone Imaging with Adaptive Optics Scanning Light
Alexander E Salmon1, Robert F Cooper2, Christopher S Langlo1
1Department of Cell Biology, Neurobiology, & Anatomy, Medical College of Wisconsin, Milwaukee, WI, USA.
Translational Vision Science & Technology
|April 11, 2017
Summary
An automated reference frame selection (ARFS) algorithm improves adaptive optics scanning light ophthalmoscope (AOSLO) video processing by selecting less distorted frames than human experts. This automation enhances clinical translation of AOSLO imaging for retinal research.
Area of Science:
- Ophthalmology
- Biomedical Imaging
- Computational Vision
Background:
- Adaptive Optics Scanning Light Ophthalmoscopy (AOSLO) enables high-resolution imaging of cone photoreceptors.
- Manual selection of reference frames for AOSLO video processing is subjective and can introduce distortions.
- Developing automated methods is crucial for consistent and efficient AOSLO image analysis.
Purpose of the Study:
- To develop and validate an Automated Reference Frame Selection (ARFS) algorithm for AOSLO video processing.
- To replace the subjective manual selection of reference frames with an objective, automated approach.
- To improve the accuracy and efficiency of cone photoreceptor imaging analysis.
Main Methods:
- Relative distortion was measured in individual frames to identify image-based motion.
- Frames were sorted into spatial clusters using image-based motion tracking.
- AOSLO images from healthy subjects and patients with retinal diseases were processed using ARFS and manual selection.
- Registration transformations to undistorted images were compared between ARFS and manual methods.
Main Results:
- ARFS significantly reduced the average transformation vector magnitude for image registration compared to manual selection (2.75 ± 1.60 pixels vs. 3.33 ± 1.61 pixels).
- ARFS rejected a substantial percentage of human-selected frames (5.16%–39.22%), indicating subjective frame selection often includes distorted images.
- Human-selected frames rarely ranked among the least distorted (2.71%–7.73% in the top 5%).
Conclusions:
- The ARFS algorithm outperforms expert observers in selecting minimally distorted reference frames for AOSLO sequences.
- Subjective assessment of image distortion by humans is challenging and less reliable than automated methods.
- ARFS facilitates a fully automated image-processing pipeline, aiding the clinical translation of AOSLO imaging.