Prognostic Analysis of Polypoidal Choroidal Vasculopathy Using an Image-Based Approach

In this paper, we rstly propose to perform prognostic analysis of polypoidal choroidal vasculopathy (PCV) using indocyanine green angiography (ICGA) sequence. Our goal is to develop a computer-aided diagnostic system which can predict the likely treatment outcome of patients with PCV based on their before-treatment ICGA sequences. In order to create a prognostic model for PCV, we utilize both the before-treatment and the aftertreatment ICGA sequences collected in the EVEREST study. By comparing the before-treatment and the after-treatment PCV region in ICGA sequences, we can generate positive and negative samples for training our prognostic model. Here, we design an 8-layer convolution neural network (CNN) and use it to serve as the prognostic model. We have conducted experiments using 17 patients cases. In particular, we perform leave-one-out cross validation so that each patient can be utilized as testing case once. Our proposed method achieves promising results on the EVEREST dataset.