| 李慕贞,罗童,陈妍,陈立业,陈伟彬.高b值弥散加权成像生境影像组学模型鉴别前列腺良、恶性病变[J].中国介入影像与治疗学,2026,23(7):409-414 |
| 高b值弥散加权成像生境影像组学模型鉴别前列腺良、恶性病变 |
| High b-value diffusion weighted imaging habitat radiomics model for differentiating benign and malignant prostate lesions |
| 投稿时间：2026-06-04 修订日期：2026-07-01 |
| DOI：10.13929/j.issn.1672-8475.2026.07.006 |
| 中文关键词: 前列腺肿瘤 弥散磁共振成像 聚类分析 影像组学 |
| 英文关键词:prostatic neoplasms diffusion magnetic resonance imaging cluster analysis radiomics |
| 基金项目:华北理工大学医工融合重点专项“揭榜挂帅”项目（ZD-YG-JBGS202509）。 |
|
| 摘要点击次数: 48 |
| 全文下载次数: 30 |
| 中文摘要: |
| 目的 观察高b值弥散加权成像（DWI）生境影像组学模型鉴别前列腺良、恶性病变的价值。方法 回顾性纳入中心1接受前列腺MR检查的406例前列腺病变，按7∶3比例随机分为训练集［n=284，前列腺癌（PCa）137例、良性病变147例］与验证集（n=122，PCa68例、良性病变54例）；以中心2的50例为外部测试集（PCa及良性病变各25例）。基于DWI（b=2 000 s/mm2）逐层手动分割病灶（MSL）区域；以3DSlicer v5.6.2的MONAI AutoSeg组件自动分割全前列腺ROI，再将其划分为DWI异常高信号区域及其邻近区域（h1）、h1亚区周围过渡区域（h2）及背景前列腺组织区域（h3）共3个生境亚区，提取并筛选单一MSL、h1、h2、h3及h1+h2+h3影像组学特征并构建相应影像组学模型或生境影像组学模型，比较各模型诊断效能。结果 生境影像组学模型鉴别训练集、验证集和外部测试集前列腺良、恶性病变的曲线下面积分别为0.917、0.911及0.870，均高于MSL、h2及h3影像组学模型（P均＜0.05）而与h1影像组学模型差异均无统计学意义（P均＞0.05）。结论 高b值DWI生境影像组学模型鉴别前列腺良、恶性病变效能较佳。 |
| 英文摘要: |
| Objective To observe the value of high b-value diffusion weighted imaging （DWI） habitat radiomics model for differentiating benign and malignant prostate lesions. Methods Totally 406 patients with prostate lesions who underwent prostate MR examinations in center 1 were retrospectively enrolled and randomly divided into training set （n=284， 137 cases of prostate cancer ［PCa］ and 147 cases of benign lesions） and validation set （n=122， 68 cases of PCa and 54 cases of benign lesions） at a ratio of 7∶3 ， while 50 cases from center 2 were taken as external test set （25 cases of PCa and 25 cases of benign lesions）. Prostate lesions were delineated layer by layer in DWI （b=2 000 s/mm²）， and manually segmented lesion （MSL） areas were obtained.Then ROI of entire prostate was automatically segmented using MONAI AutoSeg component of 3D Slicer v5.6.2 and divided into DWI abnormal high signal area and adjacent areas （h1）， the transition area around h1 subregion（h 2） and the background prostate tissue area （h3） habitat subregion， respectively. Radiomics features of MSL， h1， h2 and h3 region alone， as well as of h1+h2+h3 were extracted and screened， and relative radiomics models and habitat radiomics models were constructed. The diagnostic performance of models were evaluated and compared. Results The area under the curve of habitat radiomics model for differentiating benign and malignant prostate lesion in training set， validation set， and external test set was 0.917， 0.911 and 0.870， respectively， all higher than that of MSL， h2 and h3 radiomics models （all P＜0.05） but not significantly different compared with that of h1 radiomics model （all P＞0.05）. Conclusion High b-value DWI habitat radiomics model had good efficacy for differentiating benign and malignant prostate lesions. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
