Cancer staging is crucial in determining patient prognosis and treatment selection. The traditional interpretation of CT scans is laborious and highly susceptible to inter-observer variability, particularly in the context of Stage II, III, and IV bladder carcinoma. In this work, we present SliceNet, a deep learning slice-based framework that classifies bladder cancer into Stage II, III, and IV using CT scan images. Our model compares three state-of-the-art convolutional neural networks, ResNet18, EfficientNetB0, and ConvNeXt-Tiny, each modified to accept single-channel CT inputs. The dataset includes 2D slices of patient CT volumes obtained from the TCGA-BLCA archive. We utilized strong pre-processing, class-balancing methodologies, and patient-level stratified validation to add robustness to the model and avoid data leakage. ConvNeXt-Tiny was ultimately the best model from a validation standpoint (98.5% accuracy) with better precision, recall and generalization likely due to its transformer-based architecture. We also employed Grad-CAM visualizations to add explainability through staging-specific attention areas on the CT slices and validate clinically known markers. Our results confirm the utility of lightweight CNNs like ConvNeXt-Tiny for potential real-time, explainable bladder cancer staging and lead to future work with Vision Transformers (ViT) (ViT), volumetric learning, clinical metadata fusion, and deployment in clinical decision-support systems.