Lightweight Deep Learning for IoT-Based Medical Diagnosis: A Survey

Abstract

The rapid expansion of Internet of Things (IoT) technologies in healthcare has enabled continuous patient monitoring and data-driven clinical decision support. Deep learning models excel at extracting intricate patterns from high-dimensional medical data but face significant deployment challenges on resource-constrained IoT devices due to high computational and memory demands. This survey systematically analyzes four core strategies to adapt deep learning for edge based medical diagnosis: model pruning, quantization, knowledge distillation, and inherently lightweight architectures. For each approach, we dissect the underlying methodologies, evaluate the trade-offs between model efficiency and predictive performance, and present relevant medical case studies. We further review deployment frameworks (e.g., TensorFlow Lite, PyTorch Mobile, ONNX) that facilitate integration with IoT hardware. Evidence indicates that these lightweight techniques can substantially reduce model size and inference latency while preserving diagnostic accuracy, enabling real-time AI-powered healthcare in decentralized settings. Finally, we identify critical research challenges including energy-efficient optimization, privacy-aware model design, and end-to-end automation across software and hardware layers—and outline future directions to advance robust, efficient, and trustworthy edge AI in healthcare.