Precise and effective identification of cone photoreceptor cells is essential for comprehending retinal structure and identifying vision-related disorders. Adaptive optics scanning laser ophthalmoscopy (AOSLO) enables the capture of detailed images of the retina, allowing for the visualization of individual photoreceptor cells. Nevertheless, the process of manually identifying these cells is laborious and susceptible to mistakes. This paper presents a novel approach for automatically and without human supervision identifying cone photoreceptor cells in AOSLO pictures.

The system utilizes unsupervised learning methods to identify and separate cone cells in retinal images, without requiring labeled training data. This enhances its capacity to adapt to various types of retinal images. Important processes involve preprocessing to improve image contrast, cell segmentation using clustering techniques, and feature extraction to differentiate cones from other retinal structures. The performance of the model is assessed using a dataset of AOSLO photos, showcasing its high accuracy and resilience in many image situations, including variations in lighting and retinal health.

The results demonstrate that the suggested method substantially decreases the time needed for cell identification and attains a level of accuracy that is comparable to hand annotation. This approach has the capacity to streamline extensive retinal investigations and enhance early detection of retinal conditions such as age-related macular degeneration and retinitis pigmentosa. Potential future research could entail the incorporation of deep learning methods to enhance the accuracy of cell identification and investigate its suitability for additional types of photoreceptors.