Browser-Based Multi-Cancer Classification Framework Using Depthwise Separable Convolutions for Precision Diagnostics

Cancer is a major global health burden where early detection is paramount. While deep learning offers powerful diagnostic potential, its clinical adoption is often hin-dered by high computational costs, infrastructure demands, and data privacy con-cerns, particularly in low-resource settings. This study addresses these challenges by developing a highly accurate and universally accessible multi-cancer classification framework. We fine-tuned the Xception architecture on a comprehensive dataset of over 130,000 medical images across 26 cancer types and deployed it for client-side in-ference in a web browser using TensorFlow.js. The model’s performance was bench-marked against Visual Geometry Group–16 (VGG16) and Residual Network–50 (Res-Net50), with interpretability assessed using Grad-CAM. Our framework achieved out-standing performance with a Top-1 accuracy of 99.73%, significantly outperforming VGG16 and ResNet50. The browser-based tool enabled real-time, privacy-preserving inference, and Grad-CAM visualizations confirmed that predictions were based on clinically relevant features. This work demonstrates a viable paradigm for bridging the gap between advanced Artificial Intelligence (AI) and global health equity, offering a scalable, cost-effective, and private solution to democratize access to state-of-the-art cancer diagnostics.