Eye Image Analysis Group Rotterdam

Department                            Radiology & Nuclear Medicine/Ophthalmology

Principal investigator          Danilo Andrade De Jesus

E-mail address                      d.andradedejesus@erasmusmc.nl

Website                                     bigr.nl

 

​Fovea localization in microscopic retinal images (AO-FIO images)

Supervisor: Danilo Andrade De Jesus, d.andradedejesus@erasmusmc.nl

Adaptive Optics (AO) enables super-resolution imaging and it is used in ophthalmology to visualize retinal structures like photoreceptors. The rtx1 AO retinal camera captures 1500 x 1500 pixel images, representing 1.2mm by 1.2mm of the retina. However, cone density in the fovea (central retina) is so high that individual cells are indiscernible, leading to incorrect low-density measurements when using automatic tools. To address this, the project aims to develop an algorithm to automatically locate the fovea center and define the surrounding region where cone cells cannot be distinguished, improving accuracy, especially in patients with fixation issues.

Techniques

  • OCT-AO Image Co-localization: Optical Coherence Tomography (OCT) images will be used to precisely identify and co-localize the fovea position within Adaptive Optics (AO) images. OCT provides high-resolution cross-sectional data of the retina, enabling accurate mapping.
  • Model Training: After co-localizing the fovea, deep learning models will be trained to automatically detect the fovea’s location in AO images.
  • Algorithm Validation: The model’s performance will be validated by comparing the automatically detected fovea location with manually labeled data, focusing on accuracy and reliability.

 

​Analysis of photoreceptor density in patients using an automatic segmentation tool ​

Supervisor: Danilo Andrade De Jesus, d.andradedejesus@erasmusmc.nl

Adaptive Optics (AO) enables super-resolution imaging by correcting light aberrations and is used in ophthalmology to visualize retinal structures like photoreceptors. Previously, we developed software for locating and analyzing cones in images from the rtx1 AO retinal camera, validated in healthy subjects. Now, we aim to extend its use to patients with healthy-looking retinal areas, such as those in early stages of disease. In the AO-Vision project, we’ve acquired a dataset of over 100 patients with inherited retinal dystrophies (IRDs). The goal is to identify measurable images, analyze them, review the results, and refine the algorithm to track subtle disease progression.

Techniques

  • Image Quality Assessment: AO images from patients with inherited retinal dystrophies (IRDs) will be evaluated to identify regions with sufficient quality for analysis, excluding severely damaged areas.
  • Automatic Cone Analysis: In-house software will be used to automatically detect and analyze cones in the selected retinal regions of the IRD dataset.
  • Model Refinement: Manual review and correction of the software’s output will be performed to ensure accuracy, followed by refining the algorithm to improve performance in tracking subtle changes in retinal health.
  • Longitudinal Analysis: The refined algorithm will be used to track disease progression by analyzing how cone density and structure change over time in patients.