Lisbon · 7–10 July 2026 · past event
Treeye at AOP 2026
The VII International Conference on Applications of Optics and Photonics, at the Lisbon School of Engineering — aop2026.org (opens in a new tab).
What we presented
Telling anatomy from artefact
At AOP 2026 we took on a long-standing blind spot in anterior segment OCT: the retro-iris space, where the ciliary processes sit hidden behind a wall of iris scattering.
We showed that the signal filling that space is not uniform noise — anatomical backscatter and the iris-related artefact differ in their local intensity statistics, and can be told apart on that basis alone. This opens a path to imaging the retro-iris space with AS-OCT.
The work since has turned that finding into a method and a tool. Both were presented at ESCRS 2026, and the tool has a short demo.
-
Image analysis Conference
Anterior segment OCT image features in the retro-iris space
AOP 2026, Lisbon · doi:10.34629/ipl.isel.i-ETC.108 (opens in a new tab)
Abstract
Background. Visualization of the posterior chamber (PC) has traditionally been limited to ultrasound biomicroscopy (UBM), as iris melanin absorption typically prevents optical penetration. Long-wavelength swept-source optical coherence tomography (SS-OCT) has improved detection of deep ocular structures; however, PC imaging remains affected by signal attenuation and backscattering artifacts. This study quantified the stochastic spatial properties of SS-OCT pixels in the retro-iridial space to support robust anatomical detection.
Methods. SS-OCT B-scans from 30 eyes (Anterion, Heidelberg Engineering) were processed to characterize intensity distributions behind the iris. Nasal and temporal regions of interest (ROIs) were extracted using the iris pigment epithelium as the anterior boundary. Signal texture was quantified using local Entropy, Standard Deviation and Range of pixel intensity probability density functions (PDFs). Sampling employed a circular kernel with an anatomically matched radius. A Gaussian Mixture Model (GMM) was used to model signal components, with the optimal number of Gaussians determined by Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC). A consensus mask was generated by intersecting anatomical signal distributions and combining non-anatomical distributions. Performance was validated in 10 images using the Jaccard Index and a Symmetry Index comparing nasal and temporal morphology.
Results. BIC/AIC indicated that Entropy, Standard Deviation and Range distributions were best modeled by four, two and three Gaussian components respectively. GMM components corresponding to anatomical regions yielded: Entropy 0.66 ± 0.05, Standard Deviation 0.11 ± 0.03 and Range 0.22 ± 0.06. The consensus mask achieved a Jaccard Index of 0.54 ± 0.07. A Symmetry Index of 0.86 ± 0.05 indicated high morphological consistency between nasal and temporal PC structures.
Conclusions. SS-OCT provides sufficient spatial information to distinguish anatomical structures from noise and artifacts in the posterior chamber. Consistent pixel statistics across the population suggest that GMM-based textural analysis is a viable framework for automated detection and biometric assessment of the ciliary body and retro-iridial space.
Since then
Where this work went
Two posters at ESCRS 2026 took the method from a finding to a measurement: one showing that the enhanced images allow repeatable sulcus-to-sulcus measurement, the other testing it against ultrasound biomicroscopy.