文献速递腹部技术与安全CTMR

基于MONAI框架的3D UNet与UNet++深度神经网络用于增强腹部多器官分割

Enhanced abdominal multi-organ segmentation with 3D UNet and UNet + + deep neural networks utilizing the MONAI framework.
PubMed 文献 · Abdom Radiol (NY) · 2025-06-30

本研究在MONAI框架下实现3D UNet与UNet++架构,并采用秘书鸟优化算法(SBO)优化参数,在Pancreas-CT、Liver-CT和BTCV数据集上评估腹部多器官分割性能。3D UNet++在Pancreas-CT和Liver-CT数据集上DSC分别为95.62%和97.36%,3D UNet分别为94.54%和95.67%,BTCV数据集两模型DSC为93.42%至95.31%。

要点

  • 3D UNet++在Liver-CT数据集上DSC达97.36%,3D UNet为95.67%。
  • 3D UNet++在Pancreas-CT数据集上DSC为95.62%,3D UNet为94.54%。
  • BTCV数据集两模型DSC范围为93.42%至95.31%。

核心结论

  • 所提出的3D UNet与UNet++架构在腹部多器官分割中表现出稳健性和效率,验证了架构的准确性,为复杂腹部影像任务提供可扩展方案。

研究设计与样本

  • 研究在MONAI框架下实现3D UNet和UNet++架构,使用秘书鸟优化算法(SBO)优化参数,在Pancreas-CT、Liver-CT和BTCV多个数据集上评估性能。

诊断标准与关键数值

  • Pancreas-CT数据集:3D UNet的DSC为94.54%,3D UNet++的DSC为95.62%。
  • Liver-CT数据集:3D UNet的DSC为95.67%,3D UNet++的DSC为97.36%。
  • BTCV数据集:两模型DSC值范围为93.42%至95.31%。

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

Accurate segmentation of organs in the abdomen is a primary requirement for any medical analysis and treatment planning. In this study, we propose an approach based on 3D UNet and UNet + + architectures implemented in the MONAI framework for addressing challenges that arise due to anatomical variability, complex shape rendering of organs, and noise in CT/MRI scans. The models can analyze information in three dimensions from volumetric data, making use of skip and dense connections, and optimizing the parameters using Secretary Bird Optimization (SBO), which together help in better feature extraction and boundary delineation of the structures of interest across sets of multi-organ tissues. The developed model's performance was evaluated on multiple datasets, ranging from Pancreas-CT to Liver-CT and BTCV. The results indicated that on the Pancreas-CT dataset, a DSC of 94.54% was achieved for 3D UNet, while a slightly higher DSC of 95.62% was achieved for 3D UNet + +. Both models performed well on the Liver-CT dataset, with 3D UNet acquiring a DSC score of 95.67% and 3D UNet + + a DSC score of 97.36%. And in the case of the BTCV dataset, both models had DSC values ranging from 93.42 to 95.31%. These results demonstrate the robustness and efficiency of the models presented for clinical applications and medical research in multi-organ segmentation. This study validates the proposed architectures, underpinning and accentuating accuracy in medical imaging, creating avenues for scalable solutions for complex abdominal-imaging tasks.

来源:Abdom Radiol (NY),查看原文。版权归原作者所有。

引用

  1. Tejashwini PS, Thriveni J, Venugopal KR, et al. Enhanced abdominal multi-organ segmentation with 3D UNet and UNet + + deep neural networks utilizing the MONAI framework. Abdom Radiol (NY). 2025. PMID: 40586894. doi:10.1007/s00261-025-05041-4

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。