Effective deep learning approach to denoise optical coherence tomography images using diverse data

Retinal diseases are significant cause of visual impairment globally. In the worst case they may lead to severe vision loss or blindness. Accurate diagnosis is a key factor in right treatment planning that can stop slow disease. The examination aid Optical Coherence Tomography (OCT). OCT scans susceptible various noise effects which deteriorate their quality and as result impede analysis content. this paper, we propose novel effective method for image denoising using deep learning model trained on pairs noisy clean obtained by BM3D filtering. A comprehensive dataset 21926 scans, collected from 869 patients (1639 eyes), covering both healthy pathological cases, was used training testing proposed scheme. validated taking into account quantitative metrics concerning quality. addition, scheme evaluated analyzing impact applying it eye disease classification based Convolutional Neural Networks (CNNs) where improvement around 1-3 pp (percentage point). separate 25697 1910 (2953 eyes) purpose. conducted experiments have proved be applied preprocessing step order provide better results useful other tasks. solution much faster perform than classical filter (over ninetyfold speed-up) related methods, especially when big set images needs processed at once. Furthermore, use diverse show benefit over methods only neural network.