Adaptive Lightweight Deep Learning Models for Real-Time Medical Image Triage

Medical image triage is essential in emergency and critical care settings. Quick and accurate case prioritization can directly affect patient outcomes. However, most deep learning models used for diagnostic support are heavy and not suitable for real-time use in resource-limited places like rural hospitals, mobile medical units, and low-power clinical devices. This study presents a lightweight deep learning framework that can perform real-time triage across different medical imaging methods, such as X-ray, CT, and ultrasound. The design includes efficient feature-compression blocks, dynamic inference pathways, and a context-aware attention mechanism. These elements together reduce the computational burden while keeping diagnostic reliability.A reinforcement-learning-based adaptive controller adjusts inference paths based on image complexity and device limitations. This ensures a good balance between accuracy and speed. Experimental tests on benchmark datasets show that the model achieves significant cuts in size, latency, and energy use while still performing well compared to traditional deep learning systems. These findings demonstrate the potential of lightweight, adaptive AI models to improve access to diagnostic tools and aid timely clinical decision-making in various healthcare environments.