Cross-Domain CT Image Reconstruction with Convolution–Attention Feature Adaptation
American Journal Of Big Data · 2026-01-31 · 7 (1) · 177-184
DOI: 10.71465/ajbd3636 ↗At a glance
Journal article in American Journal Of Big Data, published online on 2026-01-31. Written by Daniel Cohen and 2 co-authors. 19 references. The full text is freely available (open access, CC BY 4.0).
Abstract (original language)
CT reconstruction models trained on data from a single scanner or acquisition protocol often show degraded performance when applied to other domains. Building on hybrid convolutionself-attention architectures exemplified by CTLformer, this paper studies cross-domain CT reconstruction using feature-level adaptation between convolutional and attention branches. The method aligns intermediate representations across domains while preserving local anatomical structures. Experiments are conducted on multi-center CT datasets comprising 62,000 slices collected from three hospitals with different scanners and dose settings. Compared with domain-agnostic CNN and transformer-based reconstruction methods, the proposed approach improves cross-domain PSNR by 1.1–1.8 dB and reduces reconstruction error variance across domains
Cite
APA
Cohen, D., Levi, Y., & Rosen, A. (2026). Cross-Domain CT Image Reconstruction with Convolution–Attention Feature Adaptation. American Journal Of Big Data, 7(1), 177-184. https://doi.org/10.71465/ajbd3636
Vancouver
Cohen D, Levi Y, Rosen A. Cross-Domain CT Image Reconstruction with Convolution–Attention Feature Adaptation. American Journal Of Big Data. 2026;7(1):177-184. doi:10.71465/ajbd3636
BibTeX
@article{Cohen2026,
author = {Cohen, Daniel and Levi, Yael and Rosen, Amit},
title = {Cross-Domain CT Image Reconstruction with Convolution–Attention Feature Adaptation},
journal = {American Journal Of Big Data},
year = {2026},
volume = {7},
number = {1},
pages = {177--184},
doi = {10.71465/ajbd3636}
}
EUVIMED