| 涂佳琪,罗中翔,刘建鹏,陈昊晴,金博,朱凤平,李郁欣,胡斌.基于MRI的深度学习联合影像组学评估中线胶质瘤H3 K27状态[J].中国医学影像技术,2024,40(6):810~814 |
| 基于MRI的深度学习联合影像组学评估中线胶质瘤H3 K27状态 |
| Deep learning combine with radiomics based on MRI for evaluating H3 K27 status of midline gliomas |
| 投稿时间：2024-02-01 修订日期：2024-04-01 |
| DOI：10.13929/j.issn.1003-3289.2024.06.004 |
| 中文关键词: 胶质瘤 深度学习 磁共振成像 影像组学 |
| 英文关键词:glioma deep learning magnetic resonance imaging radiomics |
| 基金项目:复旦大学医工结合项目(yg2023-14)。 |
|
| 摘要点击次数: 349 |
| 全文下载次数: 294 |
| 中文摘要: |
| 目的 观察基于MRI的深度学习联合影像组学评估中线胶质瘤H3 K27状态的价值。方法 回顾性收集弥漫性中线胶质瘤伴H3 K27变异(H3-DMG)患者及不伴H3 K27变异的中线胶质母细胞瘤(GBM)患者各127例,按8 ∶ 2比例将其随机分为训练集(n=204)及测试集(n=50)。基于MRI提取肿瘤U-Net神经网络视觉特征及影像组学特征,建立深度学习影像组学模型,观察其评估肿瘤H3 K27状态的价值。结果 基于训练集得出0.500为模型分类任务的安全评分划分值;以所获深度学习影像组学模型评估测试集H3-DMG和GBM H3 K27状态的中位安全评分分别为0(0,0)和0.999(0.616,1.000),前者低于后者(Z=-5.114,P＜0.001)。深度学习影像组学模型评估训练集H3 K27状态的敏感度、特异度、准确率及曲线下面积分别为93.14%、81.37%、87.25%及0.953 ,而在测试集分别为88.00%、80.00%、84.00%及0.922 。结论 基于MRI深度学习影像组学可准确评估中线胶质瘤H3 K27状态。 |
| 英文摘要: |
| Objective To observe the value of deep learning combine with radiomics based on MRI for evaluating H3 K27 status of midline gliomas. Methods Totally 127 patients with diffuse midline glioma H3 K27-altered (H3-DMG) and 127 patients with midline glioblastoma (GBM) without H3 K27 mutation were retrospectively enrolled. The patients were randomly divided into training set (n=204) and test set (n=50) at the ratio of 8:2. U-Net neural network visual and radiomics features of tumors were extracted based on MRI, and a deep learning radiomics model was established, its value for evaluating H3 K27 status was observed. Results Based on training set, 0.500 was obtained as the security score partition value for the model classification task. In test set, the median safety score of the obtained deep learning radiomics model for evaluating H3 K27 status of H3-DMG and GBM was 0 (0, 0) and 0.999 (0.616, 1.000), respectively, for the former was lower than for the latter (Z=-5.114, P＜0.001). The sensitivity, specificity, accuracy and area under the curve of deep learning radiomics model for evaluating H3 K27 status in training set was 93.14%, 81.37%, 87.25% and 0.953 (95%CI [0.923, 0.976]), respectively, while was 88.00%, 80.00%, 84.00% and 0.922 (95%CI [0.829, 0.986]) in test set, respectively. Conclusion Deep learning radiomics based on MRI could accurately evaluate H3 K27 status of midline gliomas. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
