基于1.5T自动磁共振图像分割的膝关节三维重建的可行性研究

IF 2 Q2 ORTHOPEDICS

Journal of Experimental Orthopaedics Pub Date : 2025-07-18 DOI:10.1002/jeo2.70361

Charles Pioger, Laura Marin, Yvon Gautier, Julien Cléchet, Pierre Imbert, Christian Lutz, Étienne Cavaignac, Bertrand Sonnery-Cottet

{"title":"基于1.5T自动磁共振图像分割的膝关节三维重建的可行性研究","authors":"Charles Pioger, Laura Marin, Yvon Gautier, Julien Cléchet, Pierre Imbert, Christian Lutz, Étienne Cavaignac, Bertrand Sonnery-Cottet","doi":"10.1002/jeo2.70361","DOIUrl":null,"url":null,"abstract":"<div>\n \n \n <section>\n \n <h3> Purpose</h3>\n \n <p>To validate the accuracy of three-dimensional (3D) bone and cartilage reconstructions of the distal femur and proximal tibia derived from 1.5 Tesla magnetic resonance imaging (MRI), using fully automated and semi-automated segmentation methods, compared to surface laser scanning (LS) as the reference standard.</p>\n </section>\n \n <section>\n \n <h3> Methods</h3>\n \n <p>Eleven fresh-frozen cadaveric knees were imaged using a 1.5 T MRI scanner. Manual (MS), fully automated (A), and semi-automated (SA) segmentations were performed to generate 3D models of the distal femur and proximal tibia. A transformer-based deep learning model (UNet-R) was used for automated segmentation. Laser surface scanning provided high-resolution ground-truth 3D models. Point-to-surface distances between MRI-based and LS-derived models were calculated to assess reconstruction accuracy. Bland-Altman analyses were performed to compare segmentation methods. Time to generate 3D models was recorded for each method.</p>\n </section>\n \n <section>\n \n <h3> Results</h3>\n \n <p>The mean absolute point-to-surface distance for femoral models was 1.19 mm (±0.42) for MRI A, 1.05 mm (±0.09) for MRI SA, and 0.99 mm (±0.08) for MRI MS. For tibial models, the corresponding values were 1.54 mm (±1.02), 1.03 mm (±0.17), and 0.93 mm (±0.14), respectively. MRI A showed larger variability, which required manual correction. Time analysis revealed significant efficiency gains: 27 s for MRI A, 1520 s for MRI SA, and 14,191 s for MRI MS (<i>p</i> < 0.001). Bland-Altman plots confirmed improved agreement of MRI SA with MRI MS.</p>\n </section>\n \n <section>\n \n <h3> Conclusions</h3>\n \n <p>MRI-based 3D reconstructions of the knee using a 1.5 T system and semi-automated segmentation achieved sub-millimetre accuracy comparable to manual segmentation and significantly outperformed fully automated models in precision, while substantially reducing segmentation time. These findings support the integration of AI-assisted 3D reconstruction into preoperative planning workflows for knee ligament surgery, offering a reliable, radiation-free alternative to CT-based modelling.</p>\n </section>\n \n <section>\n \n <h3> Level of Evidence</h3>\n \n <p>Level IV, controlled laboratory study.</p>\n </section>\n </div>","PeriodicalId":36909,"journal":{"name":"Journal of Experimental Orthopaedics","volume":"12 3","pages":""},"PeriodicalIF":2.0000,"publicationDate":"2025-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/jeo2.70361","citationCount":"0","resultStr":"{\"title\":\"Three-dimensional reconstruction of the knee joint based on automated 1.5T magnetic resonance image segmentation: A feasibility study\",\"authors\":\"Charles Pioger, Laura Marin, Yvon Gautier, Julien Cléchet, Pierre Imbert, Christian Lutz, Étienne Cavaignac, Bertrand Sonnery-Cottet\",\"doi\":\"10.1002/jeo2.70361\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div>\\n \\n \\n <section>\\n \\n <h3> Purpose</h3>\\n \\n <p>To validate the accuracy of three-dimensional (3D) bone and cartilage reconstructions of the distal femur and proximal tibia derived from 1.5 Tesla magnetic resonance imaging (MRI), using fully automated and semi-automated segmentation methods, compared to surface laser scanning (LS) as the reference standard.</p>\\n </section>\\n \\n <section>\\n \\n <h3> Methods</h3>\\n \\n <p>Eleven fresh-frozen cadaveric knees were imaged using a 1.5 T MRI scanner. Manual (MS), fully automated (A), and semi-automated (SA) segmentations were performed to generate 3D models of the distal femur and proximal tibia. A transformer-based deep learning model (UNet-R) was used for automated segmentation. Laser surface scanning provided high-resolution ground-truth 3D models. Point-to-surface distances between MRI-based and LS-derived models were calculated to assess reconstruction accuracy. Bland-Altman analyses were performed to compare segmentation methods. Time to generate 3D models was recorded for each method.</p>\\n </section>\\n \\n <section>\\n \\n <h3> Results</h3>\\n \\n <p>The mean absolute point-to-surface distance for femoral models was 1.19 mm (±0.42) for MRI A, 1.05 mm (±0.09) for MRI SA, and 0.99 mm (±0.08) for MRI MS. For tibial models, the corresponding values were 1.54 mm (±1.02), 1.03 mm (±0.17), and 0.93 mm (±0.14), respectively. MRI A showed larger variability, which required manual correction. Time analysis revealed significant efficiency gains: 27 s for MRI A, 1520 s for MRI SA, and 14,191 s for MRI MS (<i>p</i> < 0.001). Bland-Altman plots confirmed improved agreement of MRI SA with MRI MS.</p>\\n </section>\\n \\n <section>\\n \\n <h3> Conclusions</h3>\\n \\n <p>MRI-based 3D reconstructions of the knee using a 1.5 T system and semi-automated segmentation achieved sub-millimetre accuracy comparable to manual segmentation and significantly outperformed fully automated models in precision, while substantially reducing segmentation time. These findings support the integration of AI-assisted 3D reconstruction into preoperative planning workflows for knee ligament surgery, offering a reliable, radiation-free alternative to CT-based modelling.</p>\\n </section>\\n \\n <section>\\n \\n <h3> Level of Evidence</h3>\\n \\n <p>Level IV, controlled laboratory study.</p>\\n </section>\\n </div>\",\"PeriodicalId\":36909,\"journal\":{\"name\":\"Journal of Experimental Orthopaedics\",\"volume\":\"12 3\",\"pages\":\"\"},\"PeriodicalIF\":2.0000,\"publicationDate\":\"2025-07-18\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"https://onlinelibrary.wiley.com/doi/epdf/10.1002/jeo2.70361\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Journal of Experimental Orthopaedics\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://onlinelibrary.wiley.com/doi/10.1002/jeo2.70361\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"ORTHOPEDICS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Experimental Orthopaedics","FirstCategoryId":"1085","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.1002/jeo2.70361","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ORTHOPEDICS","Score":null,"Total":0}

引用次数: 0

摘要

目的验证1.5特斯拉磁共振成像（MRI）三维（3D）股骨远端和胫骨近端骨软骨重建的准确性，采用全自动和半自动分割方法，与表面激光扫描（LS）作为参考标准进行比较。方法采用1.5 T MRI对11例新鲜冷冻尸体膝关节进行成像。手工（MS）、全自动(A)和半自动（SA）分割生成股骨远端和胫骨近端3D模型。采用基于变压器的深度学习模型（UNet-R）进行自动分割。激光表面扫描提供了高分辨率的地面真实3D模型。计算基于mri和ls衍生模型之间的点到表面距离，以评估重建精度。进行Bland-Altman分析比较分割方法。记录每种方法生成3D模型的时间。结果股骨模型的平均绝对点面距离MRI A为1.19 mm（±0.42），MRI SA为1.05 mm（±0.09），MRI ms为0.99 mm（±0.08）。胫骨模型的相应值分别为1.54 mm（±1.02），1.03 mm（±0.17）和0.93 mm（±0.14）。MRI A显示较大的变异性，需要人工校正。时间分析显示了显著的效率提高：MRI A 27秒，MRI SA 1520秒，MRI MS 14191秒（p < 0.001）。Bland-Altman图证实了MRI SA与MRI ms的一致性。结论：使用1.5 T系统和半自动分割，基于MRI的膝关节三维重建达到了与人工分割相当的亚毫米精度，并且在精度上显著优于全自动模型，同时大大减少了分割时间。这些发现支持将人工智能辅助的3D重建整合到膝关节韧带手术的术前计划工作流程中，为基于ct的建模提供了可靠、无辐射的替代方案。证据等级四级，实验室对照研究。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

查看原文本刊更多论文

Three-dimensional reconstruction of the knee joint based on automated 1.5T magnetic resonance image segmentation: A feasibility study

Purpose

To validate the accuracy of three-dimensional (3D) bone and cartilage reconstructions of the distal femur and proximal tibia derived from 1.5 Tesla magnetic resonance imaging (MRI), using fully automated and semi-automated segmentation methods, compared to surface laser scanning (LS) as the reference standard.

Methods

Eleven fresh-frozen cadaveric knees were imaged using a 1.5 T MRI scanner. Manual (MS), fully automated (A), and semi-automated (SA) segmentations were performed to generate 3D models of the distal femur and proximal tibia. A transformer-based deep learning model (UNet-R) was used for automated segmentation. Laser surface scanning provided high-resolution ground-truth 3D models. Point-to-surface distances between MRI-based and LS-derived models were calculated to assess reconstruction accuracy. Bland-Altman analyses were performed to compare segmentation methods. Time to generate 3D models was recorded for each method.

Results

The mean absolute point-to-surface distance for femoral models was 1.19 mm (±0.42) for MRI A, 1.05 mm (±0.09) for MRI SA, and 0.99 mm (±0.08) for MRI MS. For tibial models, the corresponding values were 1.54 mm (±1.02), 1.03 mm (±0.17), and 0.93 mm (±0.14), respectively. MRI A showed larger variability, which required manual correction. Time analysis revealed significant efficiency gains: 27 s for MRI A, 1520 s for MRI SA, and 14,191 s for MRI MS (p < 0.001). Bland-Altman plots confirmed improved agreement of MRI SA with MRI MS.

Conclusions

MRI-based 3D reconstructions of the knee using a 1.5 T system and semi-automated segmentation achieved sub-millimetre accuracy comparable to manual segmentation and significantly outperformed fully automated models in precision, while substantially reducing segmentation time. These findings support the integration of AI-assisted 3D reconstruction into preoperative planning workflows for knee ligament surgery, offering a reliable, radiation-free alternative to CT-based modelling.