Cross-domain and cross-scanner generalizability

Establish the generalizability of the attention-guided multi-scale super-resolution network to 4D flow MRI acquired with other scanners, field strengths, acquisition protocols, and vascular territories.

Background

All data used in the study were acquired at a single center with one scanner and one imaging protocol, and the experiments focused on carotid arteries. Consequently, it remains unresolved whether the reported performance transfers to different acquisition settings or anatomical vascular territories.

References

Third, all data were acquired at a single center with a single scanner and protocol, so the generalizability to other scanners, field strengths, and vascular territories has not been established.

— Attention-guided super-resolution of 4D flow MRI in carotid arteries  (2609.04891 - Mokhtari et al., 4 Sep 2026) in Section Results and Discussion, limitations paragraph

The Sewer-Capsule data set is relatively small and was collected using a specific inspection platform; therefore, the reported results may not fully represent generalization across different cities, pipe materials, defect standards, imaging conditions, and robotic systems.

— Vision Transformer-Based Multi-Level Feature Fusion for Multi-Label Sewer Defect Classification  (2609.11375 - Fang et al., 10 Sep 2026) in Section 7, Limitations and Future Work

Leave-site-out validation remains an important extension for future work, as it would test whether a certificate obtained within one acquisition environment remains informative under scanner or site shift.

— When Is Accuracy Evidence? A Unified Theory of Generalisation, Validation, and Information Fusion  (2610.03465 - Gorriz, 2 Oct 2026) in Section 12, Implications for neuroimaging

Although evaluation on CAMUS-2CH, the CAMUS simulation dataset, and EchoNet-Dynamic without dataset-specific fine-tuning demonstrates transfer across the tested settings, broader generalization across institutions, scanners, acquisition protocols, image quality, and patient populations remains to be established.

— EchoDiST: Self-distillation-based joint learning for diffusion-conditioned echocardiographic myocardial motion estimation  (2610.11431 - Qi et al., 8 Oct 2026) in Discussion, paragraph beginning “Three additional limitations should be noted.”