D3T Diffusion Model: Solving 3D Incomplete-View CT Reconstruction

maier_ak · x · 2026-08-14

While CT scans provide detailed anatomical insights, they often require extensive radiation and long acquisition times. Reconstructing 3D CTs from incomplete views is an inherently "ill-posed" problem where traditional methods cause artifacts, and processing massive 3D data is computationally expensive.

To tackle this, a study published in the International Journal of Computer Vision introduces D3T (Dual-Domain Diffusion Transformer in Triplanar Latent Space). The model employs a Triplanar Vector Quantized Autoencoder in its first stage to reduce computational demands. D3T is positioned not as a replacement for full CT scans, but as a highly valuable alternative when full scans are infeasible, such as quick checks in busy operating rooms.

Original post →

More from Research

Research channel →