Object Colocalization (CPC)（对象共定位）
Object Colocalization (CPC) - User Manual
Dragonfly Prototype Apps · Object Colocalization (Spatial Relationship)...
版本 Version 1.0 · 2026-07-31
第一部分 中文手册
目录
1. 简介
2. 适用场景
3. 环境需求
1. 简介
功能：在 2–5 个已发布的标签通道（MultiROI/ROI）之间做基于对象的共定位：当一个物体的质心落入另一通道的物体内部时，即计为共定位。每一对通道双向统计（A-in-B 与 B-in-A 是两个不同的问题），并给出参考通道每个物体的组合模式（它同时与哪几个通道共定位）。可选用强度图像做质心加权（质量中心）。结果可导出 CSV/JSON 与柱状图，并可把参考通道的物体按共定位/未共定位两色发布为新 MultiROI。 本插件移植自 CPC（BSD-3-Clause，https://github.com/Jay2owe/CPC）。BSD-3 声明随包放在包根目录的 THIRD_PARTY_NOTICES.txt。 超出 CPC 的增量（已单独标注）：期望值基线。原始共定位百分比分不清“真关联”和“单纯拥挤”——目标通道占体积 40%，就会有约 40% 的质心落进去。因此用同一批质心去打随机化后的目标通道（循环平移，保持物体大小形状、只随机位置），给出富集比和置换 p 值。CPC 本身没有这项控制。
2. 适用场景
适用场景：突触与线粒体、病毒与宿主细胞器、沉淀物与晶界——任何需要回答“这些物体是不是住在那些物体里”的问题。建议始终勾选期望值基线：一个 80% 的共定位率，如果随机基线也是 78%，就什么也没说明。
3. 环境需求
环境需求：无需 GPU、联网或管理员权限，全部在 Dragonfly 内部用自带 numpy 运行。所有通道必须来自同一图像网格（形状一致）。注意：二值 ROI 会被当成一个物体，对象计数很少是你想要的；请用带标签的 MultiROI。
Part II English Manual
Contents
1. Introduction
2. Typical scenarios
3. Requirements
1. Introduction
What it does: object-based colocalization between 2-5 PUBLISHED label channels (MultiROI/ROI). An object counts as colocalized when its CENTROID falls inside an object of another channel. Every pair is scored BOTH ways (A-in-B and B-in-A are different questions), and each object of the reference channel gets its multi-target COMBINATION PATTERN (which channels it colocalizes with at once). Centroids can optionally be intensity-weighted (centre of mass). Results export to CSV/JSON plus a bar chart, and the reference objects can be published as a new MultiROI coloured colocalized / not. Ports the rule of CPC (BSD-3-Clause, https://github.com/Jay2owe/CPC). The BSD-3 notice ships in THIRD_PARTY_NOTICES.txt at the package root. ADDED beyond CPC and labelled as such: an expected-by-chance baseline. A raw coincidence percentage cannot separate association from crowding - if the target fills 40% of the volume, about 40% of centroids land in it either way - so the same centroids are re-scored against randomised copies of the target (circular shift, preserving object sizes and shapes, randomising only position), giving an ENRICHMENT ratio and a permutation p-value. CPC reports no such control.
2. Typical scenarios
Typical scenarios: synapses and mitochondria, virus particles and host organelles, precipitates and grain boundaries - anything that asks whether these objects sit inside those objects. Always tick the chance baseline: an 80% coincidence rate means nothing if the randomised baseline is 78%.
3. Requirements
Requirements: no GPU, no internet, no admin rights; runs in-process on Dragonfly's bundled numpy. All channels must come from the SAME image grid (identical shape). Note that a binary ROI counts as ONE object, which is rarely what an object count should mean - prefer a labelled MultiROI.