Abstract
Positron Emission Tomography (PET) is essential for metabolic imaging, but radiation concern motivates low-dose acquisition and detector limitations constrain spatial resolution. Existing PET enhancement methods typically treat low-dose-to-standard-dose enhancement and super-resolution as isolated tasks, and are confined to specific organs, leading to suboptimal performance and limited generalizability across heterogeneous clinical settings. To fill this gap, we propose a semantic-aware and task-controllable Denoising Diffusion Probabilistic Model (DDPM) that supports low-dose-to-standard-dose PET enhancement, super-resolution, and their joint enhancement within a unified model. Our unified network integrates anatomical priors derived from corresponding CT images with text-based acquisition metadata to enable semantic-aware PET enhancement. Task embeddings further allow the model to perform either task independently or both tasks jointly. To this end, we establish a large-scale dataset containing 2,611 paired PET/CT scans from 8 cohorts. Extensive experiments demonstrate that our model significantly improves image quality and preserves structural integrity across multiple tasks and anatomical regions, highlighting its potential for unified enhancement of degraded PET images.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_036.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=o8h0C0nD3S
BibTex
@InProceedings{CheZiy_ATaskControllable_MICCAISAT2026,
author = { Chen, Ziyi AND Li, Shengjun AND Wang, Yulin AND Sun, Kaicong AND Shen, Dinggang},
title = { { A Task-Controllable Semantic Diffusion Model for Whole-Body PET Enhancement } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17272},
month = {pending},
page = {pending}
}