Real-Time Liver Tumor Detection with a Multi-Class Ensemble Deep Learning Framework

Detecting liver tumors in large heterogeneous datasets is vital for accurate diagnosis and treatment to be performed. However, existing segmentation models struggle with multimodal tumor detection, variability in tumor shapes, over-segmentation, and noise in border regions. These issues lead to inconsistent and inaccurate results. The current study introduces a novel multiclass ensemble feature extraction and ranking-based deep learning framework to address these challenges. This framework efficiently identifies key tumor regions with a high true positive rate and maintains runtime efficiency, making it suitable for real-time liver tumor detection. Comparative evaluations using diverse liver imaging databases demonstrate the framework's superiority over existing models in terms of various classification metrics and runtime efficiency. These results highlight the framework's potential for enhancing real-time liver tumor detection applications.