Lee Eun Ji, Chang Yun-Woo, Lee Eun Hye, Cha Jang Gyu, Kim Shin Young, Choi Nami, Paek Munyoung, Darwish Omar
Department of Radiology, Soonchunhyang University Seoul Hospital, 59 Daesakwan-ro, Yongsan-ku, Seoul 04401 Korea.
Department of Radiology, Soonchunhyang University Seoul Hospital, 59 Daesakwan-ro, Yongsan-ku, Seoul 04401 Korea.
Eur J Radiol. 2025 Apr;185:111997. doi: 10.1016/j.ejrad.2025.111997. Epub 2025 Feb 12.
This study aimed to assess the image quality and the diagnostic value of deep learning reconstruction (DLR) for diffusion-weighted imaging (DWI) compared with conventional single-shot echo-planar imaging (ss-EPI) in 3 T breast MRI.
Between January and July 2023, this single-center prospective study involved patients who underwent both clinical breast MRI and additional DWIs including accelerated (fast DLR) and high-resolution (HR DLR) for the research purpose. Two radiologists independently evaluated image quality, including fat suppression homogeneity, image blurring, artifacts, and lesion conspicuity. The optimal cutoff value of the ADC value was determined based on a separate dataset comprising 98 breast lesions in 81 patients from a previous retrospective study. ADC values from 62 breast lesions (55 malignant, 7 benign) in 50 patients were analyzed to compare diagnostic performance across three DWI datasets.
The study cohort included 50 patients (median age, 55.3 years). Fast DLR and HR DLR showed significantly better image quality compared to ss-EPI (P < 0.05), with no significant difference between two DLR methods (P > 0.05). DLR protocols consistently outperform ss-EPI for reducing artifacts across all lesion types and lesion size (P < 0.05). Mean ADC values measured in the phantom and clinical images were not significantly different across DWI protocols (P > 0.05). No significant difference in the diagnostic performance with the AUC of 0.846 in ss-EPI, 0.828 in fast DLR and 0.855 in HR DLR (P > 0.05). Fast DLR showed a significantly lower standard deviation of ADC values compared to ss-EPI in malignant, mass-type lesions and those smaller than 2 cm (P < 0.05).
DLR DWI in 3T breast MRI improves image quality in both accelerated and high-resolution acquisition settings without compromising diagnostic performance. The use of DLR in DWI of breast MRI could enhance the efficiency and versatility of imaging protocols, offering significant clinical value.
本研究旨在评估在3T乳腺磁共振成像（MRI）中，深度学习重建（DLR）技术用于扩散加权成像（DWI）时的图像质量及诊断价值，并与传统的单次激发回波平面成像（ss-EPI）进行比较。
在2023年1月至7月期间，这项单中心前瞻性研究纳入了接受临床乳腺MRI检查以及为研究目的额外进行的DWI检查（包括加速版（快速DLR）和高分辨率版（HR DLR））的患者。两名放射科医生独立评估图像质量，包括脂肪抑制均匀性、图像模糊度、伪影和病变清晰度。基于先前一项回顾性研究中81例患者98个乳腺病变的单独数据集确定ADC值的最佳截断值。分析了50例患者62个乳腺病变（55个恶性，7个良性）的ADC值，以比较三个DWI数据集的诊断性能。
研究队列包括50例患者（中位年龄55.3岁）。与ss-EPI相比，快速DLR和HR DLR显示出明显更好的图像质量（P < 0.05），两种DLR方法之间无显著差异（P > 0.05）。在减少所有病变类型和病变大小的伪影方面，DLR方案始终优于ss-EPI（P < 0.05）。在不同DWI方案中，体模和临床图像中测得的平均ADC值无显著差异（P > 0.05）。ss-EPI的诊断性能AUC为0.846，快速DLR为0.828，HR DLR为0.855，无显著差异（P > 0.05）。在恶性、肿块型病变以及小于2cm的病变中，与ss-EPI相比，快速DLR显示出显著更低的ADC值标准差（P < 0.05）。
3T乳腺MRI中的DLR DWI在加速和高分辨率采集设置下均可提高图像质量，且不影响诊断性能。在乳腺MRI的DWI中使用DLR可提高成像方案的效率和通用性，具有显著的临床价值。
