Automated unsupervised identification of cone photoreceptor cells in adaptive optics scanning laser ophthalmoscope images

Abstract

Identification of cone photoreceptor cells (CPCs) is essential for the diagnosis and treatment of various retinal disorders. In the present work, a new automated unsupervised learning-based method is proposed for the identification of CPCs in adaptive optics scanning laser ophthalmoscope images. This method consists of the following steps: image denoising, CPC number estimation, bias field correction, unsupervised CPC identification, and merging close CPC identification data. By comparing our results with those obtained manually, it was found that the proposed method showed high effectiveness, with the precision, recall and F1 score values of 92.9%, 84.4% and 88.4%, respectively. Furthermore, healthy retinal AO-SLO images with the difference in CPC densities and pathological (diabetic retinopathy) AO-SLO images are processed by our method. The results demonstrated that our method exhibited a good accuracy for the identification of CPCs in diabetic retinopathy and healthy retinas.