Automated segmentation of cone cells across two adaptive optics imaging modalities, for retinal health assessment.
Adaptive optics scanning light ophthalmoscopy (AOSLO) resolves individual photoreceptors in the living human retina. Counting and measuring those cone cells by hand does not scale, and cone density is the quantity clinicians need: it is a direct indicator of retinal health and function (Shrestha et al., 2025).
Approach
The difficulty is that no single imaging modality shows everything. Confocal AOSLO images and non-confocal split-detector images expose different aspects of the same cells, so the pipeline pairs each modality with a segmentation model suited to it — StarDist for the confocal images, Cellpose for the calculated split-detector modality. Both are U-Net-based.
Segmenting rather than merely detecting cells is what makes the downstream analysis possible: from a segmentation you recover each cone’s shape and area, estimate the surrounding area occupied by rods, and from that compute cone density over the region of interest.
Result
Analyzing the same cells through two independent modalities and arriving at consistent results is the main evidence of reliability here, and what suggests the approach could be used clinically. This work was my master’s thesis at the Pattern Recognition Lab and was presented at BVM 2025.
Accurate detection and segmentation of cone cells in the retina are essential for diagnosing and managing retinal diseases. In this study, we used advanced imaging techniques, including confocal and non-confocal split detector images from adaptive optics scanning light ophthalmoscopy (AOSLO), to analyze photoreceptors for improved accuracy. Precise segmentation is crucial for understanding each cone cell’s shape, area, and distribution. It helps to estimate the surrounding areas occupied by rods, which allows the calculation of the density of cone photoreceptors in the area of interest. In turn, density is critical for evaluating overall retinal health and functionality. We explored two U-Net-based segmentation models: StarDist for confocal and Cellpose for calculated modalities. Analyzing cone cells in images from two modalities and achieving consistent results demonstrates the study’s reliability and potential for clinical application.
@inproceedings{shrestha2025cones,title={Automated Segmentation and Analysis of Cone Photoreceptors in Multimodal Adaptive Optics Imaging},author={Shrestha, Prajol and Kulyabin, Mikhail and Sindel, Aline and Pedersen, Hilde R. and Gilson, Stuart J. and Baraas, Rigmor C. and Maier, Andreas K.},booktitle={Bildverarbeitung f{\"u}r die Medizin 2025 (BVM)},series={Informatik aktuell},pages={254--259},publisher={Springer Vieweg, Wiesbaden},year={2025},doi={10.1007/978-3-658-47422-5_55},}