Ai-Powered Detection System for Multi-Liver Cancer: Integrating Genomics, Machine Learning, and Optimization Algorithms

Liver cancer remains a leading cause of global mortality, with delayed diagnosis and reliance on invasive biopsies contributing to poor patient outcomes. To address this, we developed an AI-driven diagnostic platform leveraging machine learning (Random Forest, Gradient Boosting) and optimization algorithms (Grey Wolf Optimization, Ant Colony Optimization) for non-invasive, early-stage detection. A semisupervised learning framework merged genomic datasets (770 samples, 11,150 gene features) from TCGA, Lubeck University, and GSE89377. Feature selection via XGBoost reduced the dataset to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5 0 0}$</tex> genes (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 6. 1 \%}$</tex> accuracy), followed by optimization techniques refining these to the top 5 biomarkers (COLEC10, BMP5, DNAJB14, BAP1, MSH3) while maintaining high accuracy (Gradient Boosting: 95.88%; Random Forest: 94.12%). Medical validation confirmed the biological significance of the five selected genes in liver cancer pathogenesis, underscoring their critical role in tumor progression and patient prognosis. A web application was implemented for real-time clinical decision support, enabling non-invasive analysis of genetic and clinical data. These results highlight the platform's potential to replace invasive biopsies while improving diagnostic precision. Future work will expand datasets for broader generalizability and integrate imaging data to enhance multimodal diagnostics. This study underscores the transformative role of AI and interdisciplinary collaboration in advancing liver cancer care.