Enhancing knee MRI bone marrow lesion detection with artificial intelligence: An external validation study

Research in diagnostic and interventional imaging Pub Date : 2025-08-14 DOI:10.1016/j.redii.2025.100063

Kevin Maarek , Philippine Cordelle , Tom Vesoul , Pascal Zille , Gaspard d'Assignies , Antoine Feydy , Guillaume Herpe

{"title":"Enhancing knee MRI bone marrow lesion detection with artificial intelligence: An external validation study","authors":"Kevin Maarek , Philippine Cordelle , Tom Vesoul , Pascal Zille , Gaspard d'Assignies , Antoine Feydy , Guillaume Herpe","doi":"10.1016/j.redii.2025.100063","DOIUrl":null,"url":null,"abstract":"<div><h3>Background</h3><div>Magnetic resonance imaging (MRI) is a sensitive imaging modality for identifying knee bone marrow edema, a significant biomarker in osteoarthritis and injury assessment. The precision of bone marrow edema detection is contingent upon the radiologist's expertise, and segmentation efficiency demands substantial time.</div></div><div><h3>Purpose</h3><div>This study evaluated artificial intelligence's (AI) impact on enhancing general radiologists' diagnostic accuracy for bone marrow edema detection in knee MRI.</div></div><div><h3>Materials and methods</h3><div>A multicenter, multireader, multicase methodology was used in this retrospective diagnostic study, which relied on an external dataset of 198 examinations. Mean age was 46 years with a standard deviation (SD) of 15.8 years and a female/male ratio of 49 %/51 %.</div><div>An AI algorithm from the AI solution Keros, comprising three orientation-specific 3D-UNet models, was deployed for bone marrow edema segmentation on T2/proton density with fat suppression sequences.</div><div>The ground truth was set by expert musculoskeletal radiologists.</div><div>The purpose was to externally validate the AI algorithm and compare the performance and speed of bone marrow edema identification by less experienced radiologists when using the algorithm versus not using it</div></div><div><h3>Results</h3><div>A total of 184 patients were included. With AI, readers’ sensitivity for bone marrow edema detection significantly increased by 6.1 % from 79.3 % without AI (95 % confidence interval [95 % CI]: 77.2–80.3 %) to 85.4 % (95 % CI: 84–86.2 %) with AI (<em>p</em> = 0). Specificity significantly increased by 5 % with AI assistance, reaching 93.9 % (95 % CI: 93.7–94.6 %) from 88.9 % (95 % CI: 88.6–89.4 %) (<em>p</em> = 0). Reading times were reduced by 42 % (0.66 min per exam, <em>p</em> = 3.81e-41).</div></div><div><h3>Conclusion</h3><div>AI significantly increased the sensitivity and specificity of bone marrow edema detection for general radiologists and shortened the reading process. AI-assisted detection of bone edema in the knee also opens up new perspectives for the longitudinal monitoring of patients with knee osteoarthritis.</div></div>","PeriodicalId":74676,"journal":{"name":"Research in diagnostic and interventional imaging","volume":"15 ","pages":"Article 100063"},"PeriodicalIF":0.0000,"publicationDate":"2025-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Research in diagnostic and interventional imaging","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2772652525000067","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}

引用次数: 0

Abstract

Background

Magnetic resonance imaging (MRI) is a sensitive imaging modality for identifying knee bone marrow edema, a significant biomarker in osteoarthritis and injury assessment. The precision of bone marrow edema detection is contingent upon the radiologist's expertise, and segmentation efficiency demands substantial time.

Purpose

This study evaluated artificial intelligence's (AI) impact on enhancing general radiologists' diagnostic accuracy for bone marrow edema detection in knee MRI.

Materials and methods

A multicenter, multireader, multicase methodology was used in this retrospective diagnostic study, which relied on an external dataset of 198 examinations. Mean age was 46 years with a standard deviation (SD) of 15.8 years and a female/male ratio of 49 %/51 %.

An AI algorithm from the AI solution Keros, comprising three orientation-specific 3D-UNet models, was deployed for bone marrow edema segmentation on T2/proton density with fat suppression sequences.

The ground truth was set by expert musculoskeletal radiologists.

The purpose was to externally validate the AI algorithm and compare the performance and speed of bone marrow edema identification by less experienced radiologists when using the algorithm versus not using it

Results

A total of 184 patients were included. With AI, readers’ sensitivity for bone marrow edema detection significantly increased by 6.1 % from 79.3 % without AI (95 % confidence interval [95 % CI]: 77.2–80.3 %) to 85.4 % (95 % CI: 84–86.2 %) with AI (p = 0). Specificity significantly increased by 5 % with AI assistance, reaching 93.9 % (95 % CI: 93.7–94.6 %) from 88.9 % (95 % CI: 88.6–89.4 %) (p = 0). Reading times were reduced by 42 % (0.66 min per exam, p = 3.81e-41).

Conclusion

AI significantly increased the sensitivity and specificity of bone marrow edema detection for general radiologists and shortened the reading process. AI-assisted detection of bone edema in the knee also opens up new perspectives for the longitudinal monitoring of patients with knee osteoarthritis.

查看原文本刊更多论文

人工智能增强膝关节MRI骨髓病变检测：一项外部验证研究

磁共振成像（MRI）是识别膝关节骨髓水肿的一种敏感成像方式，是骨关节炎和损伤评估的重要生物标志物。骨髓水肿检测的准确性取决于放射科医生的专业知识，分割效率需要大量的时间。目的本研究评估人工智能（AI）对提高普通放射科医生膝关节MRI骨髓水肿诊断准确性的影响。材料和方法本回顾性诊断研究采用多中心、多阅读器、多病例方法，依赖于198项检查的外部数据集。平均年龄46岁，标准差（SD） 15.8岁，男女比例为49% / 51%。来自AI解决方案Keros的AI算法，包括三个定向3D-UNet模型，采用脂肪抑制序列对T2/质子密度进行骨髓水肿分割。最基本的事实是由肌肉骨骼放射专家确定的。目的是对人工智能算法进行外部验证，并比较经验不足的放射科医生在使用该算法与不使用该算法时识别骨髓水肿的性能和速度。结果共纳入184例患者。使用人工智能后，读者对骨髓水肿检测的敏感性从未使用人工智能的79.3%（95%可信区间[95% CI]: 77.2 - 80.3%）显著提高到使用人工智能后的85.4% (95% CI: 84 - 86.2%)，提高了6.1% （p = 0）。人工智能辅助下特异性显著提高5%，从88.9% （95% CI: 88.6 - 89.4%）达到93.9% (95% CI: 93.7 - 94.6%) （p = 0）。阅读时间减少42%（每次考试0.66分钟，p = 3.81e-41）。结论人工智能显著提高了普通放射科医师骨髓水肿检测的敏感性和特异性，缩短了读取过程。人工智能辅助检测膝关节骨水肿也为膝关节骨关节炎患者的纵向监测开辟了新的视角。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Research in diagnostic and interventional imaging Radiology and Imaging

自引率

0.00%

发文量