MS-IQA: A Multi-Scale Feature Fusion Network for PET/CT Image Quality Assessment

Siqiao Li,Chen Hui,Wei Zhang,Rui Liang,Chenyue Song,Feng Jiang,Haiqi Zhu,Zhixuan Li,Hong Huang,Xiang Li
Nanjing University of Information Science and Technology; Harbin Medical University; Nanyang Technological University; Sichuan University of Science and Engineering
Medical Image Computing and Computer Assisted Intervention (MICCAI), 2025
PaperCode & Dataset
Architecture of the MS-IQA network.

MS-IQA combines complementary ResNet and Swin Transformer features across multiple scales, then predicts PET/CT quality with separate score and confidence branches.

Abstract

PET/CT quality varies with acquisition conditions and directly affects clinical interpretation. MS-IQA is a no-reference model designed for this multimodal setting. It extracts local texture with a convolutional backbone and long-range context with a Swin Transformer, aggregates features through attention-guided multi-scale fusion, and jointly estimates a quality score and its weighting. A dedicated PET/CT image-quality dataset and broad experiments support reliable prediction across scanners, doses, and anatomical regions.

Multi-Scale Dual-Backbone Design

Detailed architecture of MS-IQA.

Features from four ResNet and Swin stages are aligned and fused by attention-guided channel aggregation. The score and weight branches then combine global and scale-specific evidence.

Feature Visualization

MS-IQA feature maps on PET/CT samples.

Intermediate maps show complementary sensitivity to anatomical detail, global appearance, and quality-degraded regions before multi-scale fusion.

BibTeX

@inproceedings{li2025msiqa,
  title={{MS-IQA}: A Multi-Scale Feature Fusion Network for PET/CT Image Quality Assessment},
  author={Li, Siqiao and Hui, Chen and Zhang, Wei and Liang, Rui and Song, Chenyue and Jiang, Feng and Zhu, Haiqi and Li, Zhixuan and Huang, Hong and Li, Xiang},
  booktitle={Medical Image Computing and Computer Assisted Intervention},
  year={2025}
}