深度学习重建辅助压缩感知对乳腺T2W脂肪抑制序列图像质量的影响
Effect of deep learning reconstruction-assisted compressed sensing on the image quality of breast T2W Fat-sat sequences
黄碧云 1丁佳 1李仕广 2陈振涛 3刘世琛 3姚灵 4印宇 4段庆红5
作者信息
- 1. 贵州医科大学医学影像学院,贵州贵阳 561113
- 2. 贵阳市第二人民医院影像科,贵州贵阳 550081
- 3. 佳能医疗系统中国有限公司影像部,北京 100020
- 4. 贵州医科大学附属肿瘤医院影像科,贵州贵阳 550000
- 5. 贵州医科大学医学影像学院,贵州贵阳 561113;贵州医科大学附属肿瘤医院影像科,贵州贵阳 550000
- 折叠
摘要
目的 探讨深度学习重建(DLR)辅助常规压缩感知(CS)对乳腺磁共振T2W脂肪抑制序列图像质量的影响.方法 招募女性志愿者30名,在1.5 T磁共振仪上采用T2W脂肪预饱和(Fat-Sat)序列[加速因子(AS)为2.0、3.0及4.0]行乳腺MR扫描获得CS图像,再行DLR重建获得DLR结合CS(DLR+CS)图像,对两组图像的信噪比(SNR)和对比噪声比(CNR)及临床医生主观定性评价进行对比分析.结果 女性志愿者乳腺DLR+CS图像SNR和CNR均优于CS图像(P＜0.001),且AS为4时DLR+CS组乳腺图像SNR及CNR提升最为显著(157％及171％);女性志愿者乳腺DLR+CS图像整体图像质量、伪影、主观噪声、解剖结构显示及诊断可信度均优于CS图像(P＜0.001).结论 与常规CS图像比较,DLR辅助CS可提高乳腺T2WI Fat-Sat序列的图像质量,在较高AS条件下(3或4)依然能满足临床诊断需求.
Abstract
Objective To investigate the effect of deep learning reconstruction(DLR)-assisted compressed sensing(CS)on the image quality of breast T2W Fat-sat sequences.Methods Thirty female volunteers were recruited.Breast MR Scanning was conducted on them using a 1.5 T MR scanner with T2W fat saturation(Fat-Sat)sequences[acceleration factors(AS)2.0,3.0,and 4.0]to obtain CS images.DLR reconstruction was performed to obtain DLR combined CS(DLR+CS)images.The signal-to-noise ratio(SNR),and contrast-to-noise ratio(CNR)were compared between two groups of images.The subjective and qualitative evaluation of clinicians were assessed.Results Female volunteers had better SNR and CNR in breast DLR+CS images than CS images(P＜0.001),and the DLR+CS group showed the most significant improvement in SNR and CNR in breast images when AS was 4(157％and 171％,respectively).The overall image quality,artifacts,subjective noise,anatomical structure display and diagnostic reliability of female volunteers'breast DLR+CS images were better than those of CS images(P＜0.001).Conclusion DLR-assisted CS can improve the image quality of breast T2WI Fat Sat sequences when compared with conventional CS images.It meets clinical diagnostic needs under higher AS conditions(3 or 4).
关键词
磁共振成像/深度学习重建技术/压缩感知/脂肪抑制/乳腺/图像质量Key words
magnetic resonance imaging/deep learning reconstruction technology/compressed sensing/fat suppression/breast/image quality引用本文复制引用
基金项目
贵州省科技计划项目(黔科合基础-ZK[2023]一般006)
贵州省科技计划项目(黔科合支撑[2021]一般451)
出版年
2024
NETL
NSTL
万方数据