POLARIS, Department of Infection, Immunity & Cardiovascular Disease, The University of Sheffield, Sheffield, UK.
Department of Oncology and Metabolism, The University of Sheffield, Sheffield, UK.
J Magn Reson Imaging. 2023 Oct;58(4):1030-1044. doi: 10.1002/jmri.28643. Epub 2023 Feb 17.
Recently, deep learning via convolutional neural networks (CNNs) has largely superseded conventional methods for proton ( H)-MRI lung segmentation. However, previous deep learning studies have utilized single-center data and limited acquisition parameters.
Develop a generalizable CNN for lung segmentation in H-MRI, robust to pathology, acquisition protocol, vendor, and center.
Retrospective.
A total of 809 H-MRI scans from 258 participants with various pulmonary pathologies (median age (range): 57 (6-85); 42% females) and 31 healthy participants (median age (range): 34 (23-76); 34% females) that were split into training (593 scans (74%); 157 participants (55%)), testing (50 scans (6%); 50 participants (17%)) and external validation (164 scans (20%); 82 participants (28%)) sets.
FIELD STRENGTH/SEQUENCE: 1.5-T and 3-T/3D spoiled-gradient recalled and ultrashort echo-time H-MRI.
2D and 3D CNNs, trained on single-center, multi-sequence data, and the conventional spatial fuzzy c-means (SFCM) method were compared to manually delineated expert segmentations. Each method was validated on external data originating from several centers. Dice similarity coefficient (DSC), average boundary Hausdorff distance (Average HD), and relative error (XOR) metrics to assess segmentation performance.
Kruskal-Wallis tests assessed significances of differences between acquisitions in the testing set. Friedman tests with post hoc multiple comparisons assessed differences between the 2D CNN, 3D CNN, and SFCM. Bland-Altman analyses assessed agreement with manually derived lung volumes. A P value of <0.05 was considered statistically significant.
The 3D CNN significantly outperformed its 2D analog and SFCM, yielding a median (range) DSC of 0.961 (0.880-0.987), Average HD of 1.63 mm (0.65-5.45) and XOR of 0.079 (0.025-0.240) on the testing set and a DSC of 0.973 (0.866-0.987), Average HD of 1.11 mm (0.47-8.13) and XOR of 0.054 (0.026-0.255) on external validation data.
The 3D CNN generated accurate H-MRI lung segmentations on a heterogenous dataset, demonstrating robustness to disease pathology, sequence, vendor, and center.
Stage 1.
最近，基于卷积神经网络（CNN）的深度学习在质子（H）MRI 肺分割方面已极大地取代了传统方法。然而，之前的深度学习研究仅利用了单中心数据和有限的采集参数。
开发一种可推广的用于 H-MRI 肺分割的通用 CNN，使其对病理、采集方案、供应商和中心具有鲁棒性。
回顾性。
共有 809 例来自 258 名患有各种肺部疾病的参与者（中位年龄（范围）：57（6-85）；42%为女性）和 31 名健康参与者（中位年龄（范围）：34（23-76）；34%为女性）的 H-MRI 扫描，这些扫描分为训练集（593 例（74%）；157 名参与者（55%））、测试集（50 例（6%）；50 名参与者（17%））和外部验证集（164 例（20%）；82 名参与者（28%））。
磁场强度/序列：1.5-T 和 3-T/3D 扰相梯度回波和超短回波时间 H-MRI。
二维和三维 CNN，基于单中心、多序列数据进行训练，并与手动勾画的专家分割进行比较。每种方法均在来自多个中心的外部数据上进行验证。使用 Dice 相似系数（DSC）、平均边界 Hausdorff 距离（Average HD）和相对误差（XOR）指标评估分割性能。
Kruskal-Wallis 检验评估了测试集中采集之间差异的显著性。Friedman 检验结合事后多重比较，评估了 2D CNN、3D CNN 和 SFCM 之间的差异。Bland-Altman 分析评估了与手动生成的肺容积的一致性。P 值<0.05 被认为具有统计学意义。
3D CNN 明显优于其 2D 模拟和 SFCM，在测试集上获得了 0.961（0.880-0.987）的中位数（范围）Dice 相似系数（DSC）、1.63 mm（0.65-5.45）的平均边界 Hausdorff 距离（Average HD）和 0.079（0.025-0.240）的 XOR，在外部验证数据上获得了 0.973（0.866-0.987）的 DSC、1.11 mm（0.47-8.13）的 Average HD 和 0.054（0.026-0.255）的 XOR。
3D CNN 在异质数据集上生成了准确的 H-MRI 肺分割，证明对疾病病理、序列、供应商和中心具有鲁棒性。
4 级。
第 1 阶段。
