Application of deep learning-based compressed sensing reconstruction in three-dimensional double inversion recovery sequences
10.3969/j.issn.1002-1671.2025.06.031
- VernacularTitle:基于深度学习的压缩感知重建技术在三维双反转恢复序列中的应用
- Author:
Ziyu QIN
1
;
Meimeng ZHONG
;
Nan WANG
;
Dandan ZHENG
;
Shuo ZHANG
;
Liangjie LIN
;
Qingwei SONG
;
Chao YANG
Author Information
1. 大连医科大学附属第一医院放射科,辽宁 大连 116011
- Publication Type:Journal Article
- Keywords:
deep learning;
compressed sensing;
magnetic resonance imaging;
three-dimensional double inversion recovery
- From:
Journal of Practical Radiology
2025;41(6):1037-1041
- CountryChina
- Language:Chinese
-
Abstract:
Objective To explore the potential of CS-AI technique in accelerating cranial three-dimensional double inversion recovery(3D DIR)sequence imaging.Methods Twenty-six healthy volunteers were prospectively recruited for brain sagittal 3D DIR sequence scanning.The 3D DIR sequences were accelerated with four different acceleration factor(AF)(4,6,8,10)and reconstructed using the traditional compressed sensing(CS)algorithm and a new CS-AI algorithm.Subjective image quality was assessed by two observers using a 5-point Likert scale.Objective image quality was evaluated by calculating contrast(CN)and contrast-to-noise ratio(CNR).Firstly,using CS 4 as the standard,the optimal CS AF was derived after comparing the CN,CNR and subjective scores of CS 4 with those of CS 6,8 and 10 images in a comprehensive judgement,and then further comparing the optimal CS AF with images of CS-AI with different AF to validate the efficacy of the CS-AI,and to select the final optimal CS-AI AF.Results The comparison results between CS 4 and different CS AF indicated that CS 6 was selected as the optimal AF for CS.In further comparisons between CS and different CS-AI AF,the CS-AI technique outperformed the CS technique overall.CS-AI 8 was the maximum applicable AF.Conclusion The CS-AI is overall even better in terms of image quality with higher acceleration potential than the CS.The CS-AI 8 serves as the optimal AF and reduces scanning times by up to 50％while maintaining image quality.
