Heightmap Reconstruction of Macula on Color Fundus Images Using Conditional Generative Adversarial Networks
This work addresses a practical challenge in ophthalmology by providing additional 3D structural information from common 2D screening images, potentially aiding disease diagnosis, though it appears incremental as it builds on existing cGAN and deep supervision techniques.
The paper tackles the problem of reconstructing 3D heightmaps of the macula from 2D color fundus images, which lack depth information, by proposing a novel cGAN-based architecture that outperforms state-of-the-art methods in image translation on this task.
For screening, 3D shape of the eye retina often provides structural information and can assist ophthalmologists to diagnose diseases. However, fundus images which are one the most common screening modalities for retina diagnosis lack this information due to their 2D nature. Hence, in this work, we try to infer about this 3D information or more specifically its heights. Recent approaches have used shading information for reconstructing the heights but their output is not accurate since the utilized information is not sufficient. Additionally, other methods were dependent on the availability of more than one image of the eye which is not available in practice. In this paper, motivated by the success of Conditional Generative Adversarial Networks(cGANs) and deeply supervised networks, we propose a novel architecture for the generator which enhances the details in a sequence of steps. Comparisons on our dataset illustrate that the proposed method outperforms all of the state-of-the-art methods in image translation and medical image translation on this particular task. Additionally, clinical studies also indicate that the proposed method can provide additional information for ophthalmologists for diagnosis.