Validate in vivo generalization of the Detection Transformer

Validate whether the Detection Transformer approach for jointly estimating compartmental mean diffusivity, fractional anisotropy, principal fiber direction, and signal fraction from standard multi-shell diffusion MRI generalizes to in vivo data containing effects such as Rician noise, non-Gaussian diffusion, flow, and magnetization transfer.

Background

The study evaluates the proposed Detection Transformer exclusively on synthetic multi-compartment diffusion MRI data. The synthetic pipeline does not model several physical and biological effects expected in vivo, including Rician noise, non-Gaussian diffusion, flow, and magnetization transfer. Consequently, whether the reported reconstruction performance transfers to real in vivo measurements remains unresolved and is identified as a necessary validation step toward clinical use.

References

While the use of synthetic data allows us to objectively assess model performance, generalization to in vivo data remains to be tested, as there are several effects in vivo that are currently not modeled by our pipeline, such as Rician noise, non-Gaussian diffusion, flow, or Magnetization Transfer.

— Fiber-Resolved Microstructure Quantification from Multi-Shell Diffusion MRI using Detection Transformers  (2609.39184 - Endt et al., 30 Sep 2026) in Discussion, Section 4, p. 7