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

This research paper presents a lightweight and adaptive deep learning framework designed for real-time medical image triage. Traditional deep learning models used in diagnostic support are often computationally heavy and unsuitable for deployment in resource-limited environments such as rural hospitals, mobile medical units, and low-power clinical devices. To address this challenge, the study introduces an efficient model architecture that integrates feature-compression blocks, dynamic inference pathways, and a context-aware attention mechanism to reduce computational load while maintaining diagnostic reliability. A reinforcement-learning-based adaptive controller is incorporated to adjust inference paths based on image complexity and device constraints, ensuring an optimal balance between accuracy and processing speed. Experiments conducted on benchmark medical imaging datasets—including X-ray, CT, and ultrasound—demonstrate that the proposed system significantly reduces model size, latency, and energy consumption compared to conventional deep learning methods. The results highlight the potential of lightweight, adaptive deep learning models to enhance accessibility to diagnostic tools and support timely clinical decision-making across diverse healthcare environments