A study focused on continuous regression of Retinal Nerve Fiber Layer (RNFL) data using Multilayer Perceptrons. Through this work, we identified a critical limitation in which the model exhibited a strong tendency to fall back to predicting the mean population value.

Transitioning from grid-based architectures to non-Euclidean approaches. This ongoing project leverages Graph Neural Networks (GNNs) for continuous regression, aiming to overcome the limitations and collapsing variance seen in prior baseline models.