Enhancing Cancer Detection and Prevention using Advanced Deep Learning and AI Techniques
Infant Shervin M J, Avishikta Maity, P. Selvaraj · 2025
Accurate and early detection of kidney-related abnormalities such as cysts, stones, and tumors is vital for effective cancer prevention and treatment. The motivation behind this research stems from the urgent need to improve early detection of kidney-related abnormalities, particularly Renal Cell Carcinoma (RCC), which often progresses undetected due to subtle early-stage symptoms. Deep learning, particularly convolutional and transformer-based models, offers immense potential in medical imaging. However, to be clinically viable, such systems must not only deliver high performance but also address concerns like dataset imbalance and lack of model interpretability. This work aims to bridge these gaps by developing a robust, explainable, and efficient classification model—FusionNet—that leverages hybrid architectures and explainable AI to support radiologists in clinical decision-making and ultimately contribute to reducing mortality from late-detected renal diseases. This work also introduces a comprehensive AI-based framework that leverages deep learning techniques to enhance the classification of grayscale kidney CT images. We exploited the traditional data augmentation methods and synthetic image generation to address class imbalance using Deep Convolutional Generative Adversarial Networks (DCGANs). The proposed FusionNet architecture integrates ResNet18 for spatial feature extraction and a Vision Transformer (ViT) for capturing global contextual information, providing a powerful hybrid model for multi-class classification. The proposed system demonstrates improved accuracy and balanced performance across all four classes— Normal, Cyst, Stone, and Tumor—outperforming prior models like Kidney Ensemble-Net and transformer-only approaches. Furthermore, model explainability is ensured using Grad-CAM and LIME, helping clinicians visualize decision-making regions and enhancing trust in the AI system. This approach offers a promising step forward in AI-assisted renal cancer detection and clinical decision support.