| 洪文威,李如帅,孟庆乐,徐磊.基于扩散模型自动分割68Ga-前列腺特异性膜抗原-11 PET/MRI所示前列腺癌[J].中国医学影像技术,2025,41(2):326~330 |
| 基于扩散模型自动分割68Ga-前列腺特异性膜抗原-11 PET/MRI所示前列腺癌 |
| Automatic segmentation of prostate cancer in 68Ga-prostate specific membrane antigen-11 PET/MRI based on diffusion models |
| 投稿时间：2024-09-02 修订日期：2024-11-04 |
| DOI：10.13929/j.issn.1003-3289.2025.02.030 |
| 中文关键词: 前列腺肿瘤 正电子发射断层显像 磁共振成像 深度学习 |
| 英文关键词:prostatic neoplasms positron-emission tomography magnetic resonance imaging deep learning |
| 基金项目:江苏省医学重点学科建设单位(JSDW202247)。 |
|
| 摘要点击次数: 294 |
| 全文下载次数: 116 |
| 中文摘要: |
| 目的 观察基于扩散模型自动分割68Ga-前列腺特异性膜抗原(PSMA)-11 PET/MRI所示前列腺癌(PCa)的效果。方法 对含125例PCa患者的68Ga-PSMA-11 PET/MRI数据集进行预处理;设计基于Faster-RCNN及空间和通道重建卷积(SCConv) Diffusion级联的分割网络,其中的一级网络通过Faster-RCNN对前列腺和精囊腺进行粗定位,二级网络以基于扩散模型的SCConv Diffusion网络分割PCa;观察上述模型用于分割68Ga-PSMA-11 PET/MRI中的PCa的效果。结果 Faster-RCNN+SCConv Diffusion模型分割68Ga-PSMA-11 PET/MRI中的PCa的戴斯相似系数(DSC)、交并比(IoU)及95%豪斯多夫距离(HD)分别为0.76、0.63及20.02 mm,其性能优于nnU-Net(分别为0.73、0.62及21.20 mm)及Faster-RCNN+nnU-Net(分别为0.75、0.62及20.70 mm)模型,且其分割单发及多发PCa均更为精准,少见误分割非瘤组织。结论 基于扩散模型的Faster-RCNN+SCConv Diffusion级联网络可完整、准确地分割68Ga-PSMA-11 PET/MRI所示PCa。 |
| 英文摘要: |
| Objective To observe the effect of automatic segmentation of prostate cancer (PCa) in 68Ga-prostate specific membrane antigen (PSMA)-11 PET/MRI based on diffusion models. Methods A dataset contained 68Ga-PSMA-11 PET/MRI of 125 cases of PCa was preprocessed. Segmentation network was designed based on Faster-RCNN and spatial and channel reconstruction convolution (SCConv) Diffusion cascade, in which the first-level was used to coarsely localize the prostate and seminal vesicle glands using Faster-RCNN, and the second-level SCConv Diffusion network based on diffusion model was used to segment PCa. The effect of the above models for segmenting PCa in 68Ga-PSMA-11 PET/MRI were observed. Results The Dice similarity coefficient (DSC), intersection over union (IoU), and 95% Hausdorff distance (HD) of the Faster-RCNN+SCConv Diffusion model for segmenting PCa in 68Ga-PSMA-11 PET/MRI was 0.76, 0.63 and 20.02 mm, all superior to those of nnU-Net (0.73, 0.62 and 21.20 mm) and Faster-RCNN+nnU-Net (0.75, 0.62 and 20.70 mm) models, and the segmentation for both single and multiple PCa were all accurate, with less missegment non-tumor tissue. Conclusion Diffusion model based on Faster-RCNN+SCConv diffusion cascade network could be used to completely and accurately segment PCa in 68Ga-PSMA-11 PET/MRI. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
