Related Experiment Video
Updated: Aug 5, 2026
Quantitative Immunohistochemistry of the Cellular Microenvironment in Patient Glioblastoma Resections
Published on: July 31, 2017
Risk stratification of IDH-wildtype glioblastoma: Deep learning on histopathology images and biological
Zeyu Ma1, Zilong Wang1, Bokai Zhao2
1Department of Neurosurgery, The First Affiliated Hospital of Zhengzhou University, Zhengzhou, Henan, China.
Background:
The prognosis of IDH wild-type glioblastoma (GBM) remains poor, with limited pathological biomarkers for survival prediction. This study aimed to develop a prognostic model using deep learning-based pathological features and characterize its biological relevance.
Methods:
1038 patients with IDH wild-type GBM from the First Affiliated Hospital of Zhengzhou University were enrolled (training set: n = 839; internal validation set: n = 199), with 144 TCGA cases for external validation. 1024 features were extracted from each pathological image using deep learning techniques, and 20 prognostic features were selected via univariate Cox and least absolute shrinkage and selection operator (LASSO) regression to construct the Pathomic model. Gene Set Enrichment Analysis (GSEA) and Pearson correlation analysis were used to explore pathway associations.
Results:
The PathScore significantly stratified patients into high/low-risk groups (training set: log-rank P < 0.001; internal validation set: P = 0.023; TCGA: P = 0.017). The Pathomic-clinical model showed C-indices of 0.656, 0.672, and 0.595 in the training, internal validation, TCGA sets, respectively. GSEA identified 188 enriched pathways, primarily involving neuronal system and immune system, etc. CONCLUSIONS: The Pathomic model provides robust prognostic value for IDH wild-type GBM, linking pathological features to tumor-related biological pathways and supporting clinical decision-making.
