Automated Foveola Localization in Retinal 3D-OCT Images Using Structural Support Vector Machine Prediction (to appear in MICCAI 2012, Nice, France)

Authors
Yu-Ying Liu1, Hiroshi Ishikawa2,3, Mei Chen4, Gadi Wollstein2, Joel S. Schuman2,3, and James M. Rehg1

1School of Interactive Computing, College of Computing, Georgia Institute of Technology, Atlanta, GA 2UPMC Eye Center, University of Pittsburgh Medical Center, Pittsburgh, PA 3Department of Bioengineering, University of Pittsburgh, Pittsburgh, PA 4Intel Labs Pittsburgh, Pittsburgh, PA
Abstract
Abstract.We develop an automated method to determine the foveola location in macular 3D-OCT images in either healthy or pathological conditions. Structural Support Vector Machine (S-SVM) is trained to directly predict the location of the foveola, such that the score at the ground truth position is higher than that at any other position by a margin scaling with the associated localization loss. This S-SVM for-mulation directly minimizes the empirical risk of localization error, and makes efficient use of all available training data. It deals with the local-ization problem in a more principled way compared to the conventional binary classifier learning that uses zero-one loss and random sampling of negative examples. A total of 170 scans were collected for the experiment. Our method localized 95.1% of testing scans within the anatomical area of the foveola. Our experimental results show that the proposed method can effectively identify the location of the foveola, facilitating diagnosis around this important landmark.
Reference
Y.-Y. Liu , M. Chen, H. Ishikawa, G. Wollstein, J.S. Schuman, J. M. Rehg, "Automated Foveola Localization in Retinal 3D-OCT Images Using Structural Support Vector Machine Prediction", Intl Conf on Medical Image Computing and Computer Assisted Intervention (MICCAI), 2012 (to appear)
[Paper PDF]
Supplementary Material
Statistics of the distance between the ground truth foveola location and the scan center:


Fig. 1: Histogram of the L2 distance to the scan center on the training set (average: 5.60 pixels, only 70.8% of scans located within 6 pixels of scan center)

Fig. 2: Histogram of the L2 distance to the scan center on the testing set (average: 5.57 pixels, , only 70.4% of scans located within 6 pixels of scan center)