优化臂丛扩散张量成像:深度学习重建与同时多层加速对单次激发平面回波成像的影响
本研究在27名健康志愿者中分两阶段前瞻性评估深度学习重建对臂丛扩散张量成像质量及定量指标的影响,并探讨同时多层加速与深度学习联合的可行性。结果显示深度学习重建显著提高纤维束成像成功率、纤维数目和长度,改善读者间一致性和图像一致性,但同时多层加速会降低图像质量和纤维束成像成功率。
要点
- 深度学习重建可提高臂丛DTI纤维束成像成功率和图像一致性,Dice相似系数从0.537升至0.740。
- 同时多层加速与单次激发EPI联合在臂丛高磁敏感区域技术挑战大,会降低SNR、CNR和图像一致性。
- 深度学习重建对FA值的影响具有解剖特异性:臂丛FA降低而脊髓FA升高。
关键影像表现/征象
- 深度学习DTI在臂丛中显示显著更高的纤维束成像成功率、纤维数目和纤维长度(P<0.05)。
- 深度学习大幅改善读者间一致性和图像一致性,Dice相似系数从0.537增至0.740(P=0.006)。
- 深度学习重建导致解剖特异性FA变化:臂丛FA显著降低而脊髓FA升高。
- 同时多层加速协议较非同时多层深度学习协议纤维束成像成功率显著降低、图像质量下降(P<0.05)。
- 同时多层加速整合导致SNR和CNR降低(P<0.05),图像一致性下降,Dice相似系数从0.727降至0.486(P=0.015)。
诊断标准与关键数值
- Dice相似系数用于评估图像一致性:深度学习DTI从0.537增至0.740(P=0.006);同时多层加速整合后从0.727降至0.486(P=0.015)。
- 纤维束成像成功率、纤维数目和纤维长度在深度学习DTI中显著更高(P<0.05)。
- SNR和CNR在同时多层加速整合后显著降低(P<0.05)。
- DL12-2协议(12个扩散方向和两次b800采集)被确定为最佳时间效率方案。
临床应用与阅片要点
- 在臂丛DTI临床应用中优先考虑深度学习重建以提高纤维束成像成功率和图像一致性。
- 谨慎使用同时多层加速与单次激发EPI联合,因其在臂丛高磁敏感区域可能降低图像质量和纤维束成像成功率。
- 注意深度学习重建对FA值的影响具有解剖特异性,解读臂丛和脊髓FA值时需考虑重建方法的影响。
局限性
- 研究仅纳入健康志愿者,未在患者群体中验证。
- 样本量较小(第一阶段n=14,第二阶段n=13),可能影响统计效能和普适性。
- 研究为单中心前瞻性设计,未报告多中心外部验证。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
OBJECTIVES: This study aimed (1) to quantify the differential impact of deep learning (DL) reconstruction on brachial plexus (BP) diffusion tensor imaging (DTI) image quality and quantitative metrics compared with conventional single-shot echo-planar imaging (EPI), (2) to clarify the clinical utility and limitations of integrating simultaneous multislice (SMS) acceleration with DL DTI in the challenging BP anatomy, and (3) to determine the optimal, time-efficient DTI protocol for robust BP tractography.
MATERIALS AND METHODS: Twenty-seven healthy volunteers were prospectively enrolled (stage 1: n = 14; stage 2: n = 13). All DTI acquisitions used a single-shot EPI sequence. In the first stage, DTI with DL reconstruction was compared with a conventional non-DL DTI. In the second stage, the integration of SMS acceleration into DL DTI protocols was assessed across 5 protocol sets. Fiber tract number, tract length, tract-based DTI metrics, signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), image consistency, and interreader agreement were assessed in the BP postganglionic nerve roots and spinal cord.
RESULTS: In the first stage, DL DTI demonstrated significantly higher tractography success rates, fiber numbers, and fiber lengths in the BP ( P < 0.05). DL substantially improved interreader agreement and image consistency, with the Dice similarity coefficient (DSC) increasing from 0.537 to 0.740 ( P = 0.006). DL reconstruction resulted in changes in anatomy-specific fractional anisotropy, with a significant decrease in BP fractional anisotropy but an increase in spinal cord fractional anisotropy. In the second stage, SMS-integrated protocols produced significantly lower tractography success and degraded image quality compared with non-SMS DL protocols ( P < 0.05). SMS integration resulted in lower SNR and CNR ( P < 0.05), and reduced image consistency (DSC decreased from 0.727 to 0.486, P = 0.015). The DL12-2 protocol (12 diffusion directions and two b800 acquisitions) was identified as the optimal time-efficient solution.
CONCLUSIONS: DL reconstruction significantly improves the feasibility and reliability of BP DTI and affects quantitative metrics in an anatomy-specific manner. In contrast, the integration of SMS acceleration with single-shot EPI remains technically challenging in the high-susceptibility BP region. Our findings underscore that anatomy-specific optimization is essential for successful clinical integration of DL-enhanced DTI.
引用
- Shim E, Ham S, Kim BH, et al. Optimizing Brachial Plexus Diffusion Tensor Imaging: The Impact of Deep Learning Reconstruction and Simultaneous Multislice Acceleration on Single-shot Echo-planar Imaging. Invest Radiol. 2026. PMID: 42684110. doi:10.1097/RLI.0000000000001284
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。