深度学习MRI分类脑胶质瘤及胶质神经元肿瘤组织与分子亚型
本研究基于1844例胶质瘤及胶质神经元/神经元肿瘤患者术前多参数MRI,开发并验证了三级深度学习分类系统MRI-G/GNTs。系统采用MobileNetV2及序列前向特征选择优化多模态组合,外部测试区分六类肿瘤的AUC为0.69至0.92,准确率为0.81至0.95。
要点
- 成人型弥漫性胶质瘤与局限性星形细胞胶质瘤/GNTs的AUC达0.92,序列为T1+C、Flair、T2WI组合,鉴别价值较高。
- IDH状态及1p/19q状态分类分别用T1+C/Flair/ADC和ADC/Flair组合,外部测试AUC为0.92和0.83。
- PA与PXA鉴别用ADC/T1+C/T2WI组合,外部测试AUC达0.95,准确率最高。
核心结论
- MRI-G/GNTs系统对脑胶质瘤及GNTs的组织学和分子亚型分类具有较强能力,六类肿瘤外部测试AUC为0.69至0.92,准确率为0.81至0.95。
- 采用序列前向特征选择可优化各分类步骤的多模态MRI组合,提升分类性能。
研究设计与样本
- 研究类型为深度学习分类模型的开发与内外部验证,共纳入1844例胶质瘤或GNTs患者术前多参数MRI。
- MRI序列包括T1WI、增强后T1WI(T1+C)、T2Flair、T2WI和DWI,脑肿瘤分割模型平均Dice为0.92。
- 分类框架采用MobileNetV2构建三级分类系统MRI-G/GNTs,并利用序列前向特征选择优化多模态网络组合。
- 六类分类为:少突胶质细胞瘤IDH突变伴1p/19q共缺失、星形细胞瘤IDH突变、胶质母细胞瘤IDH野生型、毛细胞星形细胞瘤、多形性黄色星形细胞瘤、GNTs。
关键影像表现/征象
- 区分成人型弥漫性胶质瘤与局限性星形细胞胶质瘤/GNTs的外部测试AUC为0.92,采用T1+C、Flair、T2WI组合。
- 区分成人型弥漫性胶质瘤IDH状态的外部测试AUC为0.92,采用T1+C、Flair、ADC组合。
- 区分IDH突变弥漫性胶质瘤1p/19q状态的外部测试AUC为0.83,采用ADC、Flair组合。
- 区分局限性星形细胞胶质瘤与GNTs的外部测试AUC为0.84,采用ADC、T1+C组合。
- 区分毛细胞星形细胞瘤与多形性黄色星形细胞瘤的外部测试AUC为0.95,采用ADC、T1+C、T2WI组合。
诊断标准与关键数值
- 六类肿瘤分类外部测试AUC分别为0.92、0.81、0.87、0.82、0.94、0.69,准确率分别为0.88、0.88、0.89、0.81、0.91、0.95。
- 区分成人型弥漫性胶质瘤与局限性星形细胞胶质瘤/GNTs的外部测试AUC为0.92,优于部分单序列或固定组合结果。
- 区分IDH状态外部测试AUC为0.92,区分1p/19q状态外部测试AUC为0.83,均基于优化后的多模态组合。
- 区分PA与PXA外部测试AUC达0.95,为六类区分中性能最高的指标。
临床应用与阅片要点
- 术前多参数MRI可辅助预测胶质瘤IDH突变及1p/19q共缺失状态,有助于报告提示分子分型。
- 阅片时T1+C、Flair、ADC的组合在弥漫性胶质瘤分级及分子标记预测中价值较高,可优先关注这些序列。
- 局限性星形细胞胶质瘤与GNTs鉴别需结合ADC和T1+C序列,PA与PXA鉴别可参考ADC、T1+C、T2WI特征。
- 深度学习输出可整合入结构化虚拟病理报告,辅助临床决策,但不能替代组织病理和分子检测。
局限性
- 原文未报告研究设计的单中心或多中心性质,亦未报告内外部数据集的具体来源和样本分配细节。
- 原文未提供各分类阈值、置信区间及统计显著性检验结果。
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
PURPOSE: To develop a deep learning classification system for integrated histology and molecular subtyping of gliomas and glioneuronal/neuronal tumors (GNTs).
MATERIALS AND METHODS: Preoperative multi-parametric MRI data from 1,844 patients with gliomas or GNTs, encompassing T1WI, post-contrast T1WI (T1(+C)), T2Flair, T2WI, and DWI, were analyzed. Subjects were classified into six categories: (1) oligodendroglioma, IDH-mutant, and 1p/19q-codeleted; (2) astrocytoma, IDH-mutant; (3) glioblastoma, IDH-wildtype; (4) pilocytic astrocytoma (PA); (5) pleomorphic xanthoastrocytoma (PXA); and (6) GNTs. A three-level classification system (MRI-G/GNTs) using MobileNetV2 was trained and validated in internal and external datasets. Sequential forward feature selection (SFFS) was utilized to optimize multi-modal network combinations. Furthermore, the MRI-G/GNTs outputs were integrated into structured virtual pathology reports.
RESULTS: The average Dice for the brain tumor segmentation model was 0.92. Using SFFS, we identified optimal MRI sequence combinations per classification step. The external tests AUCs reached 0.92 for distinguishing adult-type diffuse gliomas from circumscribed astrocytic gliomas/GNTs (T1(+C), Flair, and T2WI combination), 0.92 for distinguishing IDH status for adult-type diffuse glioma (T1(+C)/Flair/ADC), 0.83 for distinguishing 1p/19q status for IDH-mutant diffuse glioma (ADC/Flair), 0.84 for circumscribed astrocytic gliomas vs GNTs (ADC/T1(+C)), 0.95 for PA vs PXA (ADC/T1(+C)/T2WI). Based on the automated prediction pipeline, the integrated classification performance of the MRI-G/GNTs system achieved AUCs of 0.92, 0.81, 0.87, 0.82, 0.94, and 0.69, with accuracy of 0.88, 0.88, 0.89, 0.81, 0.91, and 0.95 across the above six tumor categories.
CONCLUSIONS: The MRI-G/GNTs system demonstrates strong capability in distinguishing both histological and molecular subtypes of brain gliomas and GNTs.
KEY POINTS: Question How about the deep learning model based on MRI for integrated histology and molecular classification of brain GNTs, and can it be improved? Findings The MRI-G/GNTs system achieved AUCs of 0.69-0.92 with accuracy of 0.81-0.95 for distinguishing six tumor categories of GNTs. Clinical relevance The three-level deep learning classification system (MRI-G/GNTs) based on optimized multimodal MRI integration firstly achieves comprehensive histology and molecular subtype classification of brain GNTs.
引用
- Wu, Sun, Bao, et al. Classification of histology and molecular subtypes of brain gliomas and glioneuronal and neuronal tumors using a deep learning approach. Eur Radiol. 2026. PMID: 42763331. doi:10.1007/s00330-026-12857-1
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。