空间图谱加权损失集成用于全身PSMA PET/CT转移分割
原标题：Spatial Atlas-Informed Loss Ensembles for Whole-Body PSMA PET/CT Metastasis Segmentation
AI 摘要
研究提出SAILS方法，将基于375例专家标注扫描构建的全身体素级前列腺癌转移图谱整合进nnU-Net训练，通过图谱加权损失函数惩罚漏检和误报，推理时无需图谱或配准。在AutoPET III和Deep-PSMA两个数据集上，SAILS提升了病灶级敏感性和高风险骨转移分割Dice，但假阳性体积有所增加。该工作表明空间先验可改善转移灶检测，同时揭示了灵敏度与假阳性之间的权衡。
正文节选
Spatial Atlas-Informed Loss Ensembles for Whole-Body PSMA PET/CT Metastasis Segmentation Abstract Background: Existing whole-body PSMA PET/CT prostate metastasis segmentation methods generally rely on images alone, without explicitly incorporating where metastases tend to occur. We investigate whether a population metastasis atlas can improve lesion detection when integrated into model training. Methods: We constructed voxel-wise mean and standard deviation maps from 375 expert-annotated trainin