Cause of modality-specific impact of spatial dimensionality

Investigate the cause of the contrasting impact of spatial model dimensionality on synthetic CT generation performance, specifically why 2D models outperform 2.5D and 3D models for MRI-to-CT synthesis while 3D models outperform 2D and 2.5D models for CBCT-to-CT synthesis, and determine the modality-specific factors responsible for these differences.

Background

The challenge found opposing trends: for MRI-to-CT synthesis, 2D models significantly outperformed 2.5D and 3D patch-based approaches, whereas for CBCT-to-CT synthesis, 3D patch-based models significantly outperformed 2D and 2.5D models.

This contradicts parts of prior literature and suggests modality-specific factors may drive these differences. The authors explicitly state they did not identify the cause, pointing to an unresolved question requiring targeted investigation.