Xu Houshi, Fan Zhen, Jiang Shan, Sun Maoyuan, Chai Huihui, Zhu Ruize, Liu Xiaoyu, Wang Yue, Chen Jiawen, Wei Junji, Mao Ying, Shi Zhifeng
Department of Neurosurgery Huashan Hospital Shanghai Medical College Fudan University Shanghai China.
Research Unit of New Technologies of Micro-Endoscopy Combination in Skull Base Surgery (2018RU008) Chinese Academy of Medical Sciences and Peking Union Medical College Beijing China.
MedComm (2020). 2025 Apr 22;6(5):e70138. doi: 10.1002/mco2.70138. eCollection 2025 May.
Glioma subtyping is crucial for treatment decisions, but traditional approaches often fail to capture tumor heterogeneity. This study proposes a novel framework integrating multiplex immunohistochemistry (mIHC) and machine learning for glioma subtyping and prognosis prediction. 185 patient samples from the Huashan hospital cohort were stained using a multi-label mIHC panel and analyzed with an AI-based auto-scanning system to calculate cell ratios and determine the proportion of positive tumor cells for various markers. Patients were divided into two cohorts (training: = 111, testing: = 74), and a machine learning model was then developed and validated for subtype classification and prognosis prediction. The framework identified two distinct glioma subtypes with significant differences in prognosis, clinical characteristics, and molecular profiles. The high-risk subtype, associated with older age, poorer outcomes, astrocytoma/glioblastoma, higher tumor grades, elevated mesenchymal scores, and an inhibitory immune microenvironment, exhibited IDH wild-type, 1p19q non-codeletion, and MGMT promoter unmethylation, suggesting chemotherapy resistance. Conversely, the low-risk subtype, characterized by younger age, better prognosis, astrocytoma/oligodendroglioma, lower tumor grades, and favorable molecular profiles (IDH mutation, 1p19q codeletion, MGMT promoter methylation), indicated chemotherapy sensitivity. The mIHC-based framework enables rapid glioma classification, facilitating tailored treatment strategies and accurate prognosis prediction, potentially improving patient management and outcomes.
胶质瘤亚型分类对于治疗决策至关重要，但传统方法往往无法捕捉肿瘤异质性。本研究提出了一种整合多重免疫组化（mIHC）和机器学习的新框架，用于胶质瘤亚型分类和预后预测。使用多标记mIHC面板对来自华山医院队列的185例患者样本进行染色，并通过基于人工智能的自动扫描系统进行分析，以计算细胞比例并确定各种标志物的阳性肿瘤细胞比例。将患者分为两个队列（训练组：n = 111，测试组：n = 74），然后开发并验证了一种机器学习模型用于亚型分类和预后预测。该框架识别出两种不同的胶质瘤亚型，它们在预后、临床特征和分子谱方面存在显著差异。高风险亚型与年龄较大、预后较差、星形细胞瘤/胶质母细胞瘤、肿瘤分级较高、间充质评分升高以及抑制性免疫微环境相关，表现为异柠檬酸脱氢酶（IDH）野生型、1p19q非共缺失和O6-甲基鸟嘌呤-DNA甲基转移酶（MGMT）启动子未甲基化，提示化疗耐药。相反，低风险亚型的特征是年龄较小、预后较好、星形细胞瘤/少突胶质细胞瘤、肿瘤分级较低以及良好的分子谱（IDH突变、1p19q共缺失、MGMT启动子甲基化），表明对化疗敏感。基于mIHC的框架能够实现快速的胶质瘤分类，有助于制定个性化的治疗策略和准确的预后预测，可能改善患者管理和治疗结果。
