High-energy field-of-view completion in dual-source dual-energy CT using measurement-conditioned frequency-decoupled learning

This study proposes a measurement-conditioned framework for high-energy field-of-view (FOV) completion in dual-source dual-energy computed tomography (DECT). The model uses a full-FOV low-energy image, a measured central high-energy seed, and its support mask, while hard data consistency preserves the measured seed. A frequency-decoupled objective combines image-domain reconstruction, high-frequency adversarial supervision, and wavelet-based low-frequency attenuation constraints. Because true high-energy references are unavailable outside the physical high-energy FOV, a controlled seed–halo setting was used by withholding an annular region within the measured FOV for paired supervision and evaluation. Leave-one-patient-out experiments were conducted on thoracoabdominal DECT data from three patients. In the held-out halo, the proposed method achieved the lowest MAE and RMSE of 28.56 ± 6.23 HU and 40.86 ± 8.32 HU, respectively, and the highest SSIM of 0.9420 ± 0.0220 among the evaluated methods. It also more closely reproduced tissue-specific HU values and local intensity variation. In virtual monoenergetic imaging, it achieved the lowest MAE at all evaluated energies. Although true outer-FOV accuracy could not be directly validated, the results support the feasibility of the proposed approach for high-energy FOV completion and downstream spectral imaging.