Deep Classification Framework For Real-Time Imbalanced Liver Tumor Databases Using Combining Probabilistic Segmentation With Ensemble Feature Extraction

As liver tumor image datasets continue to grow, conventional prediction methods encounter difficulties due to significant imbalances between majority and minority classes, as well as the presence of noise. While 3D convolution effectively handles spatial data, it requires substantial GPU resources, whereas 2D convolution is limited in its ability to fully capture the third dimension. Challenges like missing data, noise, and class imbalance adversely affect classification performance, highlighting the critical role of data quality. This study introduces an enhanced ensemble classification model that utilizes k-joint probabilistic segmentation for detecting liver tumors, relying on medical imaging techniques such as CT for feature extraction and segmentation. By incorporating novel filtering and ranking approaches, the model achieves superior recall, precision, and AUC in comparison to current develops.