A Survey of Methods and Architectures for Segmentation and Classification of Renal Tumors Using EfficientNet-Based U-Net and Epistemic Neural Networks

Abstract

Renal tumor detection and segmentation are crucial for early diagnosis and effective treatment planning in kidney cancer. With the increasing availability of medical imaging data, deep learning techniques have significantly improved the accuracy and efficiency of automated tumor analysis. This survey explores recent methods and architectures for renal tumor segmentation and classification, focusing on EfficientNet-based U-Net models and epistemic neural networks for uncertainty estimation. EfficientNet enhances feature extraction through compound scaling, while U-Net provides precise localization via its encoder–decoder structure. The integration of epistemic neural networks further improves model reliability by quantifying prediction uncertainty, which is essential in clinical applications. Recent advancements between 2020 and 2023 demonstrate the effectiveness of hybrid architectures, attention mechanisms, and multi-stage frameworks in achieving high segmentation accuracy. Benchmark datasets such as KiTS19 and KiTS21 have been widely used for evaluation. Despite notable progress, challenges such as data imbalance, computational complexity, and limited generalization remain. This survey provides a comprehensive overview of existing techniques, comparative analysis of methods, and insights into future research directions for developing robust and clinically deployable renal tumor analysis systems.