AI-based analysis of MGMT immunohistochemical expression in glioma

Background and Objective: Glioma is the most common primary malignant tumor of the central nervous system. The expression status of O⁶-methylguanine-DNA methyltransferase (MGMT) is closely associated with patient prognosis and treatment response. This study aims to develop a deep learning-based artificial intelligence method (utilizing the CLAM framework combined with UNI feature extraction) for the accurate interpretation of MGMT immunohistochemical expression in glioma, thereby providing an auxiliary tool for evaluating the efficacy of temozolomide and exploring its potential combined therapeutic strategies with immunotherapy.Methods: This study included 270 glioma cases. MGMT immunohistochemically stained slides from formalin-fixed paraffin-embedded (FFPE) specimens of these cases were analyzed using two classification methods: Method 1 (Positive/Negative) and Method 2 (High/Low expression, with a threshold of 2%). Additionally, the correlation between MGMT expression and programmed death-ligand 1 (PD-L1) as well as microsatellite instability (MSI) was preliminarily explored in 12 glioma samples.Results: Method 1 achieved a classification accuracy of 0.85 and an AUC value of 0.856, while Method 2 showed both an accuracy and an AUC of 0.77. The performance of Method 1 was superior to that of Method 2. In the MGMT low-expression group, an upward trend in PD-L1 expression was observed.Conclusion: The AI-based interpretation method demonstrates accuracy superior to manual assessment by reducing subjectivity and interference from confounding factors. This technique not only helps alleviate the workload in clinical pathological diagnosis but also provides reliable evidence for the precise diagnosis and treatment of glioma.