LDCT-to-SDCT as a Bridge Problem: Single-Step Residual Endpoint Flow Matching for Real-Time Denoising

Abstract

Diffusion models achieve high reconstruction quality in lowdose computed tomography (LDCT), but their iterative sampling trajectories impose substantial computational costs. Unlike unconditional generation, paired LDCT reconstruction starts from an image that already contains the anatomy and spatial structure of the standard-dose CT (SDCT) target; reconstruction primarily requires correcting doserelated noise and artifacts. We therefore introduce Residual Endpoint Flow Matching (REFM), an LDCT reconstruction method that learns to transport an LDCT image directly toward its paired SDCT endpoint rather than defining a noise-to-image trajectory. REFM predicts the residual velocity along linear interpolations between both images and supports single-step and multi-step reconstruction using the same trained network. We evaluate five model capacities using 1 to 50 Euler steps against deterministic U-Net and diffusion-based baselines. Across all REFM variants, one-step inference consistently provides the highest reconstruction quality. On the TCIA validation set, REFM Base achieves 50.98 dB PSNR and 0.9865 SSIM at 94.54 fps, compared with 50.92 dB, 0.9847, and 9.26 fps for DDPM-10. REFM Small retains 50.71 dB while increasing throughput to 198.56 fps. Without fine-tuning, REFM Base also matches the 25-step DDPM baseline on the external Mayo Clinic dataset, although DDPM remains stronger on synthetically degraded CRLM images. Thus, our results show that exploiting paired anatomical correspondence enables diffusion-level LDCT reconstruction with a single step reconstruction.

Contemporary cardiometabolic and inflammation targeting therapies reduce residual cardiovascular risk in an endpoint harmonized meta analysis
Musa Bin Bashir, Chen Chen, Azhgool Mohammad +2
Scientific Reports15 Oct 2026Peer-reviewedCC BY 4.0Full textSurgery & anaesthesiaCrossref
BLINATUMOMAB AS A BRIDGE THERAPY TO HEMATOPOIETIC STEM CELL TRANSPLANTATION IN A PATIENT WITH B-CELL ACUTE LYMPHOBLASTIC LEUKEMIA REFRACTORY TO MULTIPLE CHEMOTHERAPY REGIMENS: ACHIEVEMENT OF MEASURABLE RESIDUAL DISEASE NEGATIVITY AND IMPACT ON OUTCOME
Elder Trindade DAMASCENO, Leonardo Augusto Toledo KISTER, Luana Sávia Santos SILVA +10
Hematology, Transfusion and Cell Therapy23 Oct 2026Peer-reviewedCC BYFull textEpidemiology & public healthDOAJ
Cumulative endpoint and reconstructed 500-m energy-system contributions from individualized pacing-matched trials in 2000-m rowing ergometry
Jiashi Lin, Xiaoning Ma, Caoan Yuan +1
Frontiers in physiology16 Oct 2026Peer-reviewedCC BYFull textGeneral medicineEurope PMC
inClinico Agent for clinical trial endpoint prediction and prospective evaluation
J. Chen, C. Duan, W. Gou +9
medRxiv13 Oct 2026PreprintAll rights reservedFull textEpidemiology & public healthmedRxiv
Treatment of ANCA-Associated Vasculitis With Avacopan: Results of the ADVOCATE Trial After Readjudication of the Primary Endpoint
D. Jayne, P. A. Merkel, X. Tang +6
medRxiv16 Sep 2026PreprintLicence not statedFull textGeneral medicinemedRxiv

Cite

BibTeX RIS

BibTeX

@misc{sotta2026ldcttosdct,
  title = {{LDCT-to-SDCT as a Bridge Problem: Single-Step Residual Endpoint Flow Matching for Real-Time Denoising}},
  author = {T. dela Sotta and J. M. Saavedra and V. Chang and A. Xavier and H. Henriquez and Y. Orellana and J. Curimil},
  howpublished = {medRxiv},
  year = {2026},
  month = aug,
  doi = {10.64898/2026.08.27.26361520},
  url = {https://www.medrxiv.org/content/10.64898/2026.08.27.26361520v1},
  note = {CC BY},
}

RIS

TY  - UNPB
TI  - LDCT-to-SDCT as a Bridge Problem: Single-Step Residual Endpoint Flow Matching for Real-Time Denoising
AU  - T. dela Sotta
AU  - J. M. Saavedra
AU  - V. Chang
AU  - A. Xavier
AU  - H. Henriquez
AU  - Y. Orellana
AU  - J. Curimil
PY  - 2026
DA  - 2026/08/27
PB  - medRxiv
DO  - 10.64898/2026.08.27.26361520
UR  - https://www.medrxiv.org/content/10.64898/2026.08.27.26361520v1
AB  - Diffusion models achieve high reconstruction quality in lowdose computed tomography (LDCT), but their iterative sampling trajectories impose substantial computational costs. Unlike unconditional generation, paired LDCT reconstruction starts from an image that already contains the anatomy and spatial structure of the standard-dose CT (SDCT) target; reconstruction primarily requires correcting doserelated noise and artifacts. We therefore introduce Residual Endpoint Flow Matching (REFM), an LDCT reconstruction method that learns to transport an LDCT image directly toward its paired SDCT endpoint rather than defining a noise-to-image trajectory. REFM predicts the residual velocity along linear interpolations between both images and supports single-step and multi-step reconstruction using the same trained network. We evaluate five model capacities using 1 to 50 Euler steps against deterministic U-Net and diffusion-based baselines. Across all REFM variants, one-step inference consistently provides the highest reconstruction quality. On the TCIA validation set, REFM Base achieves 50.98 dB PSNR and 0.9865 SSIM at 94.54 fps, compared with 50.92 dB, 0.9847, and 9.26 fps for DDPM-10. REFM Small retains 50.71 dB while increasing throughput to 198.56 fps. Without fine-tuning, REFM Base also matches the 25-step DDPM baseline on the external Mayo Clinic dataset, although DDPM remains stronger on synthetically degraded CRLM images. Thus, our results show that exploiting paired anatomical correspondence enables diffusion-level LDCT reconstruction with a single step reconstruction.
ER  -

Abstract

Related

Cite