PO.PR01.03 · 预防研究
面向结直肠息肉诊断检测开发的高多重空间图谱分析与复杂度降低转化框架
A translational framework for high-plex spatial profiling and complexity reduction toward diagnostic assay development in colorectal polyps
该海报暂无可下载的资料
AACR 官方页面
作者与单位 Authors & Affiliations
摘要 Abstract
中文摘要
背景：
对结直肠息肉进行准确的风险分层对于减少不必要的随访监测、同时确保高风险患者获得及时干预至关重要。病理学工作流程依赖于H&E切片的形态学，而多重免疫荧光（mIF）等新兴免疫图谱分析技术虽然可提供更深层的生物学分辨率，但对于常规临床部署而言往往过于复杂且成本高昂。为此，我们提出一种诊断检测开发框架，将高多重空间图谱分析与计算性复杂度降低策略相整合，以推导出临床上切实可行的生物标志物组合。
方法：
我们设计了一个20-plex的mIF panel，用于表征结直肠息肉微环境内的免疫细胞群体及空间相互作用。对相应的H&E全切片图像进行分析，采用基于面积的模型和计算形态学描述符提取上皮、间质及结构特征。这些多模态数据被整合到一个统一的预测建模流程中，用于将患者分层为低风险组和风险升高组。为支持向可部署检测的转化，我们实施了一个复杂度降低框架，纳入迭代式特征选择、冗余消除、模型剪枝以及可用于检测的标志物子集的模拟。
结果：
初始数据集包含200张mIF和H&E切片，用于模型微调、生物标志物特征提取以及免疫与形态学特征的初步整合。早期阶段基于mIF的模型捕获了>10个免疫细胞群体，区分了上皮亚型，并定位了关键的微环境相互作用。基于H&E的模型识别了结直肠区室、间质-上皮组织结构、炎症模式以及与异型增生相关的特征。这一基础工作实现了特征集的优化、模型稳定性的评估，以及多模态融合策略的建立，为后续的预测建模和检测简化提供指导，这些内容稍后将在更大规模的约1000例样本队列中加以验证。
结论：
我们提出了一个可扩展的框架，将高多重mIF发现与基于H&E的计算形态学相统一，以支持结直肠息肉风险分层的生物标志物识别、特征降低及诊断检测开发。该平台为即将开展的临床验证以及在结直肠监测项目中的部署奠定了基础。
查看英文原文 English abstract
Background:
Accurate risk stratification of colorectal polyps is essential for reducing unnecessary surveillance while ensuring that high-risk patients receive timely intervention. Pathology workflows rely on morphology from H&E slides, while emerging immune-profiling techniques such as multiplex immunofluorescence (mIF) offer deeper biological resolution but are often too complex and costly for routine clinical deployment. To address this, we propose a diagnostic-assay development framework that integrates high-plex spatial profiling with computational complexity-reduction strategies to derive a clinically practical biomarker panel.
Methods:
We designed a 20-plex mIF panel to characterize immune cell populations and spatial interactions within colorectal polyp microenvironments. Corresponding H&E whole-slide images were analyzed to extract epithelial, stromal, and architectural features using area-based models and computational morphology descriptors. These multimodal data were integrated into a unified predictive modeling pipeline for stratifying patients into low- and elevated-risk groups. To support translation into a deployable assay, we implemented a complexity-reduction framework incorporating iterative feature selection, redundancy elimination, model pruning, and simulation of assay-ready marker subsets.
Results:
An initial dataset of 200 mIF and H&E slides was used for model fine-tuning, biomarker feature extraction, and preliminary integration of immune and morphological signatures. Early-stage mIF-based models captured >10 immune cell populations, distinguished epithelial subtypes, and localized key microenvironmental interactions. H&E-based models identified colorectal compartments, stromal-epithelial organization, inflammatory patterns, and dysplasia-related features. This groundwork enabled refinement of feature sets, assessment of model stability, and establishment of the multimodal fusion strategy guiding downstream predictive modeling and assay simplification later to be verified on a larger ~1000 sample cohort.
Conclusions:
We present a scalable framework that unifies high-plex mIF discovery with H&E-based computational morphology to support biomarker identification, feature reduction, and diagnostic assay development for colorectal polyp risk stratification. This platform provides the foundation for forthcoming clinical validation and deployment within colorectal surveillance programs.
利益披露 Disclosure
E. Markovits,
Nucleai Employment.
G. P. Lynch, None.
O. Rimer-Cohen,
Nucleai Employment.
A. Lynch, None.
M. Azulay,
Nucleai Employment, Stock Option.
L. Sakhneny,
Nucleai Employment.
L. Irvine, None.
K. Bloom,
Nucleai Employment.
G. Greene,
Nucleai Employment.