Title: Integrative explainable artificial intelligence and metabolic biomarker analysis for early detection of oral squamous cell carcinoma
Abstract:
Background: Early detection of oral squamous cell carcinoma remains a major clinical challenge due to the subjective nature of histopathological grading of oral epithelial dysplasia. Recent advances in artificial intelligence have enabled automated analysis of digital pathology images, yet the biological mechanisms underlying AI-detected morphological patterns remain largely unexplored.
Objective: This study aimed to develop an integrative framework combining explainable multi-instance learning based histopathological grading with molecular and bioinformatic analysis to identify biological pathways associated with oral cancer progression.
Methods: Digital histopathology images of oral epithelial dysplasia were analyzed using an explainable deep learning framework based on multi-instance learning to automatically classify dysplasia severity and highlight diagnostically relevant tissue regions. To investigate the molecular basis of these morphological alterations, quantitative RT-PCR was performed to evaluate the expression of the metabolic gene CBS and signaling regulators RGS2 and RGS7 in oral squamous cell carcinoma, dental pulp, and normal tissue samples. Complementary bioinformatic analyses including gene ontology, pathway enrichment, and protein interaction network mapping were conducted to elucidate functional roles of these genes in tumor biology.
Results: The explainable AI model successfully identified histological patterns associated with epithelial dysplasia and malignant transformation. RT-PCR analysis demonstrated significant differential expression of CBS and RGS genes in oral cancer tissues compared with normal controls. Bioinformatic pathway analysis revealed involvement of these genes in metabolic regulation, redox homeostasis, and G-protein signaling pathways linked to tumor development.
Conclusion: This study demonstrates a multimodal strategy integrating computational pathology with molecular biomarker validation to improve understanding of oral cancer progression. The findings highlight the potential of combining explainable AI-based dysplasia grading with gene-level analysis to advance precision diagnostics in oral oncology.
