SALIENT: Frequency-Aware Paired Diffusion for Controllable Long-Tail CT Detection
Yifan Li, Mehrdad Salimitari, Taiyu Zhang, Guang Li, David Dreizin · 2026
Read SALIENT: Frequency-Aware Paired Diffusion for Controllable Long-Tail CT Detection by Yifan Li, Mehrdad Salimitari, Taiyu Zhang, Guang Li, David Dreizin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-23447.txt.
Description
Yifan Li and four co-authors present SALIENT, a system for creating synthetic CT scans to help detect rare lesions. Rare lesions in whole-body CT are hard to spot because they are scarce and tiny relative to scan volume, so precision collapses even when overall scores look strong.
SALIENT runs diffusion over discrete wavelet coefficients rather than raw pixels, splitting low-frequency brightness from high-frequency detail. Learnable frequency-aware goals separate target traits from background traits, and a 3D VAE makes varied lesion masks with paired slice labels. Realism improves sharply: MS-SSIM rises from zero point six three to zero point eight three and FID falls from one hundred eighteen point four to forty-six point five. Training with SALIENT scans also lifts long-tail detection, with the largest AUPRC gains at low prevalence.
Catalog record
| CREATOR | Yifan Li, Mehrdad Salimitari, Taiyu Zhang, Guang Li, David Dreizin |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-23447