Ocular Fundus Imaging: From Structure to Function P. Serranho 1,* , C.Maduro 1,2 , T.Santos 1,2 , J.Cunha-Vaz 1,2 , R.Bernardes 1,2 1 IBILI-Institute of Biomedical Research on Light and Image Faculty of Medicine, University of Coimbra 2 AIBILI - Association for Innovation and Biomedical Research on Light and Image Coimbra, Portugal * pserranho@fmed.uc.pt Adérito Araújo 3 , Silvia Barbeiro 3 3 Mathematics Department Faculty of Science and Technology, University of Coimbra Coimbra, Portugal Abstract— Imaging the ocular fundus, namely the retina, to detect and/or monitor changes over time from the healthy condition is of fundamental importance to assess onset and disease progression and is a valuable tool to understand the basic mechanisms of ocular diseases. Current trends point to the need for less or non-invasive approaches, to the need for detailed (higher spatial and temporal resolution) imaging systems and to the quantification as opposed to qualitative classification of any findings. In this work we present a snapshot of our research by presenting two examples of technical development aiming to obtain structural and function information from the human retina, in vivo, using non-invasive techniques, namely optical coherence tomography imaging. Based on our experience and developed work, we are now starting to bridge the gap to brain imaging as the eye is only the starting point of vision. Index Terms—Optical Coherence Tomography, despeckling, image segmentation . I. INTRODUCTION Vision is one of the most valued senses in humans and its loss or degradation is therefore of paramount importance. The eye, where it all starts, plays in this way a fundamental role in vision by representing the front-end part of a complex and still not fully understood system. Within our research group we are focused in the posterior segment of the human eye, the retina, and in developing procedures/imaging modalities able to provide early quantitative indications on any alterations occurring in the ocular fundus. These alterations can be either structural, functional or both and we aim to understand how they correlate with each other and with disease onset/progression, eg. diabetes, one of the main causes of vision loss in developed countries. Moreover, we aim to move from qualitative to quantitative imaging, from invasive to non-invasive systems and from documentation to diagnosis imaging. In this way, we have been actively contributing to knowledge in field of diabetic retinopathy, mainly, by developing functional imaging modalities, unifying different sources into multimodal imaging and quantifying traditional imaging modalities. We are currently deeply involved in getting functional information from structural imaging modalities as the wide spread optical coherence tomography (OCT) imaging modality. In the rest of this paper we will briefly step by recent contributions in relation with the OCT, namely the development of a complex diffusion despeckling filter and work on automated segmentation of retinal structures, fundamental steps towards functional imaging through OCT. OCT is a non-invasive imaging modality with an increasing number of applications and it is becoming an essential tool in ophthalmology [1] allowing in vivo high- resolution cross-sectional imaging of the retinal tissue. However, as any imaging technique that bases its image formation on coherent waves, OCT images suffer from speckle noise, which reduces its quality. Speckle noise creates a grainy appearance that can mask diagnostically significant image features (small or low reflectivity features) and reduce the accuracy of segmentation and pattern recognition algorithms [2-5]. In this work we present some recent developments to overcome these problems. We suggest an improved adaptive complex diffusion despeckling filter that suppresses speckle noise while preserving features within the tissue. This can be seen as a first step towards the segmentation of the image. We also present the foundations of a fully automatic computational light support vector machine (SVM) based segmentation algorithm for segmentation of the retina in an OCT image. The classification for the training set is obtained by gradient methods, namely the mean-shift algorithm. No manual input is needed, opposite to other segmentation methods based on classification [6, 7]. II. METHODS A. Improved adaptive complex diffusion despeckling filter The general nonlinear complex diffusion filtering (NCDF) method [4, 5] looks for the solution of the complex diffusion equation This work is funded by PTDC/SAU-BEB/103151/2008 and program COMPETE (FCOMP-01-0124 FEDER-010930).