Retinal Images Segmentation Lua Ngo January 29th 2018 Bionics and Photonics Lab. at Korea Univ. Multi-level deep neural network for retinal fundus image blood vessel segmentation • A max-resizing multi-level network is used for segmenting the blood vessel in retinal fundus image Fig. 2 Evaluation accuracy and loss Fig. 1 a Model architecture, b Max-resizing illustration Fig. 3 Predicted results Bionics and Photonics Lab. at Korea Univ. Deep neural network regression for OCT image segmentation • Each boundary pixel is predicted as a regression analysis applied deep neural network by 3 fe atures extracted, adaptive normalized intensity score, intensity and vertical gradient Fig. 7 Segmentation procedure. Fig. 9 Features extraction illustration. Fig. 8 Training and testing process. Bionics and Photonics Lab. at Korea Univ. Thank you! Bionics and Photonics Lab. at Korea Univ.