基于人工智能的飞行时间磁共振血管造影上脑动静脉畸形巢的自动分割
本研究纳入92例同时接受TOF-MRA和DSA检查的脑动静脉畸形患者,由两名神经外科医生在TOF-MRA图像上手动分割畸形巢作为参考标准,构建基于U-Net的人工智能模型实现畸形巢自动检测与分割。AI分割与手动分割的平均畸形巢体积分别为5.427±4.996 mL和4.824±4.567 mL,差异无统计学意义(P=0.693),测试集DSC为0.754±0.074,两者体积线性相关系数为0.988(p<0.001),提示AI分割与手动分割具有中等程度一致性,在术前规划、疗效评估、风险分层和随访中具有较大潜力。
要点
- AI模型可在TOF-MRA上自动分割脑动静脉畸形巢,与手动分割体积差异无统计学意义(P=0.693)。
- 测试集DSC为0.754±0.074,精确率0.713±0.102,召回率0.816±0.098,与手动分割呈中等程度一致。
- AI与手动分割的畸形巢体积线性相关系数为0.988(p<0.001),两者体积高度相关。
核心结论
- AI分割模型的表现与手动分割呈中等程度一致,在术前规划、疗效评估、风险分层和随访等临床场景中具有较大潜力。
研究设计与样本
- 本研究纳入92例同时接受TOF-MRA和DSA检查的脑动静脉畸形患者。
- 由两名神经外科医生在TOF-MRA图像上手动分割畸形巢,作为参考标准(ground-truth reference)。
- 构建基于U-Net的人工智能模型,用于TOF-MRA图像上畸形巢的自动检测与分割。
诊断标准与关键数值
- AI分割模型与参考标准的平均畸形巢体积分别为5.427±4.996 mL和4.824±4.567 mL,两组平均体积差为0.603±1.514 mL,差异无统计学意义(P=0.693)。
- 测试集DSC为0.754±0.074,精确率为0.713±0.102,召回率为0.816±0.098。
- AI分割与参考标准的畸形巢体积线性相关系数为0.988(p<0.001)。
临床应用与阅片要点
- 原文未据此给出推荐。
局限性
- 原文未报告局限性。
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
OBJECTIVE: Accurate nidus segmentation and quantification have long been challenging but important tasks in the clinical management of Cerebral Arteriovenous Malformation (CAVM). However, there are still dilemmas in nidus segmentation, such as difficulty defining the demarcation of the nidus, observer-dependent variation and time consumption. The aim of this study isto develop an artificial intelligence model to automatically segment the nidus on Time-Of-Flight Magnetic Resonance Angiography (TOF-MRA) images.
METHODS: A total of 92patients with CAVM who underwent both TOF-MRA and DSA examinations were enrolled. Two neurosurgeonsmanually segmented the nidusonTOF-MRA images,which were regarded as theground-truth reference. AU-Net-basedAImodelwascreatedfor automatic nidus detectionand segmentationonTOF-MRA images.
RESULTS: The meannidus volumes of the AI segmentationmodeland the ground truthwere 5.427 ± 4.996 and 4.824 ± 4.567 mL,respectively. The meandifference in the nidus volume between the two groups was0.603 ± 1.514 mL,which wasnot statisticallysignificant (P = 0.693). The DSC,precision and recallofthe testset were 0.754 ± 0.074, 0.713 ± 0.102 and 0.816 ± 0.098, respectively. The linear correlation coefficient of the nidus volume betweenthesetwo groupswas 0.988, p < 0.001.
CONCLUSION: The performance of the AI segmentationmodel is moderate consistent with that of manual segmentation. This AI model has great potential in clinical settings, such as preoperative planning, treatment efficacy evaluation, riskstratification and follow-up.
引用
- Dong M, Xiang S, Hong T, et al. Artificial intelligence-based automatic nidus segmentation of cerebral arteriovenous malformation on time-of-flight magnetic resonance angiography. Eur J Radiol. 2024. PMID: 39002268. doi:10.1016/j.ejrad.2024.111572
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。