Improving Kidney Tumor Detection Accuracy Using Hybrid U-Net Segmentation

Abstract

Kidney cancer stands as a significant factor in cancer-related mortality, highlighting the critical importance of early and precise tumor detection This study introduces a computer-aided approach using the KiTS19 dataset and a hybrid U-Net architecture. Manual tumor segmentation is resource-intensive and prone to errors. Leveraging the hybrid U-Net, known for its proficiency in medical image analysis, we achieve precise tumor identification. Our method involves initial kidney and tumor segmentation in high-resolution CT images, followed by region of interest (ROI) generation and benign/malignant tumor classification. The assessment conducted on the KiTS19 dataset demonstrates encouraging outcomes, with Dice coefficients of 0.974 for kidney segmentation and 0.818 for tumor segmentation, accompanied by a tumor classification accuracy rate of 94.3%.The hybrid U-Net’s advanced feature extraction and spatial context awareness contribute to these outcomes. By streamlining diagnosis, our approach has the potential to significantly improve patient outcomes. The use of the KiTS19 dataset ensures robustness across various clinical cases and imaging modalities. This method represents a valuable advancement in computer-aided kidney tumor detection, promising to enhance patient care.