Deep learning-based segmentation of retinal blood vessels in fluorescein angiography

Retinal eye diseases refer to conditions affecting the retina that may lead to impaired visual function or even blindness. Identifying these diseases at their earliest phase is necessary to prevent any form of visual impairment. The blood vessel segmentation from fundus color images is one of the most important aspects of the diagnosis of retinal diseases such as diabetic retinopathy, hypertensive retinopathy, and glaucoma. Here, a new blood-signal detection method is presented, which combines multiscale analysis derived by the stationary wavelet transform and a fully complex multiscale neural network. Basically, the proposed method adjusts all the differences to the vessel width and the retina orientation. To augment the data and improve prediction accuracy, rotation operations were applied at least once across the layers during the training phase. The performance of the proposed method on the three different datasets is better than that of the current methods. Besides, the proposed method is also stable, i.e., it produces consistent results across different training datasets and inter-rater variabilities. So, this method can be practically used anywhere.