LG-nnU-net用于MRI肛门外括约肌多标签分割:肛瘘患者的定量评估
单中心回顾性研究纳入272例肛瘘患者冠状位T2加权MRI,按8:2分为训练集218例和测试集54例,由三名腹部放射科医师手动标注肛提肌、耻骨直肠肌及肛门外括约肌浅部和皮下部分。LG-nnU-net在全部肌肉结构及括约肌复合体分割上DSC均优于ResU-net、DenseU-net和U-net++,括约肌复合体整体DSC为81.34%。
要点
- LG-nnU-net对肛提肌、耻骨直肠肌、肛门外括约肌浅部和皮下部分的分割DSC分别为76.89%、76.71%、68.82%和79.89%。
- 括约肌复合体整体多结构分割DSC为81.34%、HD 10.48 mm、ASSD 0.46 mm,优于所有对比模型。
- 作者指出结果提示可用于定量分析和术前评估,但对高位/低位肛瘘分类的临床影响尚未验证。
核心结论
- LG-nnU-net框架能够对MRI上的肛门外括约肌各亚结构进行准确、稳健的分割。
- 结果提示该框架在定量分析和术前评估方面具有潜在用途,但其对高位/低位肛瘘分类的临床影响尚未验证。
研究设计与样本
- 单中心回顾性研究,经机构伦理委员会批准,纳入2017年1月至2024年12月间272例诊断为肛瘘并接受冠状位T2加权MRI的患者。
- 数据集按8:2比例分为训练集218例和测试集54例。
- 由三名经验丰富的腹部放射科医师手动标注肛提肌(LAM)、耻骨直肠肌(PRM)及肛门外括约肌(EAS)浅部和皮下部分。
- 采用优化nnU-net架构(LG-nnU-net),包含非对称编码器扩展、组归一化、多尺度特征聚合和深度监督;与ResU-net、DenseU-net和U-net++比较。
诊断标准与关键数值
- LG-nnU-net分割肛提肌:DSC 76.89%(95%CI 75.1-78.7%),HD 9.45 mm,ASSD 0.74 mm。
- LG-nnU-net分割耻骨直肠肌:DSC 76.71%(95%CI 74.8-78.6%),HD 11.11 mm,ASSD 0.71 mm。
- LG-nnU-net分割肛门外括约肌浅部:DSC 68.82%(95%CI 66.3-71.4%),HD 9.17 mm,ASSD 0.92 mm;分割皮下部分:DSC 79.89%(95%CI 77.9-81.9%),HD 8.06 mm,ASSD 0.54 mm。
- LG-nnU-net分割联合高层结构(LAM+PRM)DSC 79.94%(95%CI 78.0-81.9%),联合低层结构(浅部+皮下EAS)DSC 80.32%(95%CI 78.5-82.1%)。
- 括约肌复合体整体多结构分割DSC 81.34%(95%CI 79.6-83.1%),HD 10.48 mm,ASSD 0.46 mm,优于所有对比模型。
局限性
- 作者指出该框架对高位/低位肛瘘分类的临床影响尚未验证,需要进一步的多中心验证和前瞻性结局研究。
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 and evaluate a novel deep learning-based segmentation framework (LG-nnU-net) for multilabel segmentation of anal sphincter substructures on MRI, aimed at providing robust quantitative anatomical information without implying operative validation for clinical classification improvement.
METHODS: This single-center retrospective study, approved by the institutional ethics committee, included 272 patients diagnosed with anal fistula who underwent coronal T2-weighted MRI between January 2017 and December 2024. The dataset was divided into training and testing subsets in an 8:2 ratio, comprising 218 and 54 patients, respectively. Manual annotations of the levator ani muscle (LAM), puborectalis muscle (PRM), and superficial and subcutaneous parts of the external anal sphincter (EAS) were performed by three experienced abdominal radiologists. An optimized nnU-net architecture (LG-nnU-net) was implemented with asymmetric encoder expansion, Group Normalization, multi-scale feature aggregation, and deep supervision. Segmentation performance was compared to ResU-net, DenseU-net, and U-net++ using Dice similarity coefficient (DSC), Hausdorff distance (HD), and average symmetric surface distance (ASSD) for individual muscles, combined anatomical structures for quantitative evaluation only, and the entire sphincter complex.
RESULTS: LG-nnU-net achieved superior segmentation accuracy for all individual muscle structures: LAM DSC 76.89 % (95 % CI: 75.1-78.7 %), HD 9.45 mm, ASSD 0.74 mm; PRM DSC 76.71 % (95 % CI: 74.8-78.6 %), HD 11.11 mm, ASSD 0.71 mm; superficial EAS DSC 68.82 % (95 % CI: 66.3-71.4 %), HD 9.17 mm, ASSD 0.92 mm; subcutaneous EAS DSC 79.89 % (95 % CI: 77.9-81.9 %), HD 8.06 mm, ASSD 0.54 mm. For combined high-level structures (LAM + PRM) and low-level structures (superficial + subcutaneous EAS), DSCs were 79.94 % (95 % CI: 78.0-81.9 %) and 80.32 % (95 % CI: 78.5-82.1 %), respectively. The overall multi-structure segmentation of the anal sphincter complex achieved a DSC of 81.34 % (95 % CI: 79.6-83.1 %), HD 10.48 mm, and ASSD 0.46 mm, surpassing all comparison models.
CONCLUSION: The LG-nnU-net framework enables accurate, robust segmentation of anal sphincter substructures on MRI. These results indicate potential utility for quantitative analysis and preoperative assessment, but the clinical impact on high/low fistula classification remains unverified, requiring further multi-center validation and prospective outcome studies.
引用
- Liu X, Ren H, Lv J, et al. LG-nnU-net for multilabel anal sphincter segmentation on MRI: quantitative evaluation in patients with anal fistula. Eur J Radiol. 2025. PMID: 41232256. doi:10.1016/j.ejrad.2025.112509
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。