Explainable Deep CNN Ensemble for Multi-Class Kidney Lesion Classification in CT Imaging
Year: 2026
Type: conference-paper
Abstract: Accurate detection of renal abnormalities is critical for preventing chronic kidney disease progression and reducing diagnostic delays. However, existing deep learning models for renal CT analysis often suffer from overfitting, class imbalance sensitivity, and limited interpretability. This paper proposes Ensemble-ASNet, an optimized weighted soft-voting ensemble that integrates three high-performing pre-trained Convolutional Neural Network (CNN) backbones, such as InceptionV3, VGG16, and MobileNetV2, to improve robustness and discriminative capability for four-class renal CT classification (Cyst, Normal, Stone, and Tumor). Optimal ensemble weights are learned via constrained validation-based optimization. Experimental results demonstrate that Ensemble-ASNet consistently outperforms individual models and recent CNN, transformer, and hybrid-based approaches, achieving 99.42% accuracy with equally strong precision, recall, and F1-score values. Grad-CAM++ visualizations confirm attention to clinically relevant anatomical regions, enhancing interpretability. These results highlight Ensemble-ASNet as an effective and reliable framework for automated renal disease diagnosis using CT images.
Institution: Sylhet Engineering College
Language: English
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7204469232