Evaluation of 2D and 3D Deep Learning Approaches For Predicting Visual Acuity Following Surgery for Idiopathic Full-Thickness Macular Holes In Spectral Domain Optical Coherence Tomography Images

In this work, we compared the performance of 2D and 3D versions of four state-of-the-art deep learning neural networks on predicting visual acuity following surgery for idiopathic full-thickness macular holes using an image dataset of spectral-domain optical coherence tomography (OCT) scans. To make this study more comparable, using the same dataset revealed the differences between 2D and 3D versions of deep learning neural networks. Based on our results, 3D networks generally outperformed the 2D networks in R-squared and Pearson correlation coefficient; however, they fell behind in mean absolute error. 3D networks also come with the sacrifice of significantly more computational complexity.