Fundus images play a key role in clinical diagnosis and are especially critical for the diagnosis of retinal diseases. However, current fundus images are usually two-dimensional
and lack three-dimensional depth information, which poses a limitation for physicians to fully understand patients’ ocular diseases. For diseases that require depth information on the fundus surface such as glaucoma [MacCormick et al., 2019], commonly used diagnostic methods such as fundus OCT often do not provide sufficient 3D information, and these methods do not include background parameters regarding fundus photography information.
In addition, since the fundus is located inside the eye, acquiring its corresponding 3D image is often not easy, especially when using a mobile camera like the remidio.
To address this challenge, this study worked on developing a method that can estimate 3D surface contours from monocular fundus images to provide more information about the surface structure of the eye. We created a dataset containing fundus OCT images and their corresponding 3D truth values, named 3D-CSCR.Based on this dataset, we developed a method capable of constructing corresponding 3D models from monocular fundus images and constructed an average template of the fundus 3D model to provide generic structural features.
The results of our study show that our method has made significant progress in providing depth information, which provides ophthalmologists with more comprehensive
image information and thus contributes to a more accurate diagnosis of ocular diseases, especially diseases like glaucoma, which require depth information. In addition, our
study provides new perspectives for improving ophthalmic medical diagnosis and lays a solid foundation for research and development in the field of 3D reconstruction based on monocular fundus images.