Tariq Humera, Hadjiiski Lubomir, Malyarenko Dariya, Talpaz Moshe, Pettit Kristen, Luker Gary D, Ross Brian D, Chenevert Thomas L
Department of Radiology, University of Michigan, Ann Arbor, MI, United States.
Department of Internal Medicine-Hematology/Oncology, University of Michigan, Ann Arbor, MI, United States.
Front Oncol. 2025 May 23;15:1498832. doi: 10.3389/fonc.2025.1498832. eCollection 2025.
PURPOSE: To automate bone marrow segmentation within pelvic bones in quantitative fat MRI of myelofibrosis (MF) patients using deep-learning (DL) U-Net models. METHODS: Automated segmentation of bone marrow (BM) was evaluated for four U-Net models: 2D U-Net, 2D attention U-Net (2D A-U-Net), 3D U-Net and 3D attention U-Net (3D A-U-Net). An experienced annotator performed the delineation on in-phase (IP) pelvic MRI slices to mark the boundaries of BM regions within two pelvic bones: proximal femur and posterior ilium. The dataset comprising volumetric images of 58 MF patients was split into 32 training, 6 validation and 20 test sub-sets. Model performance was assessed using conventional metrics: average Jaccard Index (AJI), average Volume Error (AVE), average Hausdorff Distance (AHD), and average Volume Intersection Ratio (VIR). Iterative model optimization was performed based on maximizing validation sub-set AJI. Wilcoxon's rank sum test with Bonferroni corrected significance threshold of p<0.003 was used to compare DL segmentation models for test sub-set. RESULTS: 2D segmentation models performed best for iliac BM with achieved scores of 95-96% for the VIR and 87-88% for AJI agreement with expert annotations on the test set. Similar performance was observed for femoral BM segmentation with slightly better VIR but worse AJI agreement for U-Net (94% and 86%) versus A-U-Net (92% and 87%). 2D models also exhibited lower AVE variability (8-9%) and ilium AHD (16 mm). The 3D segmentation models have shown marginally higher errors (AHD of 19-20 mm for ilium and 10-12% AVE-SD for both bones) and generally lower agreement scores (VIR of 91-93% for ilium and 89-91% for femur with 85-86% AJI).Pairwise comparison across four U-Nets for three metrics (AHD, AJI, AVE) showed that AJI and AHD performance was not significantly different for 3D U-Net versus 3D A-U-Net and for 2D U-Net versus 2D A-U-Net. Except for AVE, for majority of performance metric comparisons 2D versus 3D model differences were significant in both bones (p<0.001). CONCLUSION: All four tested U-Net models effectively automated BM segmentation in pelvic MRI of MF patients. The 2D A-U-Net was found best overall for BM segmentation in both femur and ilium.
目的：使用深度学习（DL）U-Net模型，在骨髓纤维化（MF）患者的定量脂肪MRI中实现骨盆骨内骨髓分割的自动化。 方法：对四种U-Net模型评估骨髓（BM）的自动分割：二维U-Net、二维注意力U-Net（2D A-U-Net）、三维U-Net和三维注意力U-Net（3D A-U-Net）。一位经验丰富的注释者在同相（IP）骨盆MRI切片上进行描绘，以标记两个骨盆骨（股骨近端和髂骨后部）内BM区域的边界。将包含58例MF患者体积图像的数据集分为32个训练子集、6个验证子集和20个测试子集。使用常规指标评估模型性能：平均杰卡德指数（AJI）、平均体积误差（AVE）、平均豪斯多夫距离（AHD）和平均体积交集比（VIR）。基于最大化验证子集AJI进行迭代模型优化。使用威尔科克森秩和检验，校正后的显著性阈值p<0.003，比较测试子集的DL分割模型。 结果：二维分割模型对髂骨BM的表现最佳，在测试集上VIR得分达到95-96%，与专家注释的AJI一致性为87-88%。股骨BM分割也观察到类似的性能，U-Net（94%和86%）的VIR略好，但AJI一致性比A-U-Net（92%和87%）差。二维模型的AVE变异性（8-9%）和髂骨AHD（16毫米）也较低。三维分割模型显示出略高的误差（髂骨AHD为19-20毫米，两块骨的AVE-SD为10-12%），总体一致性得分较低（髂骨VIR为91-93%，股骨VIR为89-91%，AJI为85-86%）。对四个U-Net在三个指标（AHD、AJI、AVE）上进行成对比较，结果显示三维U-Net与三维A-U-Net以及二维U-Net与二维A-U-Net的AJI和AHD性能无显著差异。除AVE外，在大多数性能指标比较中，二维和三维模型在两块骨中的差异均显著（p<0.001）。 结论：所有四个测试的U-Net模型都有效地实现了MF患者骨盆MRI中BM分割的自动化。发现二维A-U-Net在股骨和髂骨的BM分割中总体表现最佳。
Med Phys. 2019-2-28
J Med Imaging (Bellingham). 2023-11
Transl Vis Sci Technol. 2023-11-1
IEEE Trans Pattern Anal Mach Intell. 2024-12
Med Image Anal. 2024-5
Comput Struct Biotechnol J. 2023-12-27
Nat Commun. 2022-8-17
Zotero 插件
