Pathomics-based machine learning model predicts interferon-gamma expression in head and neck squamous cell carcinoma patients

Abstract

Introduction: Interferon-gamma (IFNG) plays a key role in immune responses in head and neck squamous cell carcinoma (HNSCC) and impacts the effectiveness of immune checkpoint inhibitors. This study developed a machine learning model that leverages pathomics to forecast IFNG expression based on histopathological images while thoroughly examining the tumor immune microenvironment in HNSCC.

Methods: The analysis involved 271 cases from The Cancer Genome Atlas (TCGA)-HNSCC, with validation using 71 patients from the Hospital. Significant links were found between IFNG expression, clinical features, and survival outcomes. For histopathological image processing, tumor regions were segmented using the OTSU algorithm, and 1,488 features were extracted with PyRadiomics. A feature selection strategy that integrated minimum Redundancy Maximum Relevance (mRMR) and Recursive Feature Elimination (RFE) pinpointed 30 essential features, which were then employed to construct a Gradient Boosting Machine (GBM) predictive.

Results: This model demonstrated strong performance in predicting survival, with AUC values of 0.836 in the TCGA training set, 0.753 in the validation set, and 0.740 in the hospital dataset. Gene set variation analysis revealed distinct pathway activation patterns among PS subgroups, indicating that pathways associated with immune evasion were more prominent in patients with high PS. Survival analysis substantiated that patients exhibiting elevated IFNG expression experienced a longer median survival.

Discussion: In conclusion, the study shows that pathomics from histopathological images can predict IFNG expression. The correlation between IFNG and pathomics provides a valuable biomarker framework for elucidating the pathophysiology of HNSCC and may inform personalized therapeutic strategies through non-invasive characterization of the immune microenvironment.

Keywords: IFNg; biomarkers; head and neck squamous cell carcinoma; machine learning; pathomics.