A collaborative inference strategy for medical image diagnosis in mobile edge computing environment.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Mobile Edge Computing

Background:

• Mobile medical image analysis in edge computing offers convenience but faces DNN performance, energy, and communication challenges.
• Existing collaborative approaches exhibit unstable performance and inefficient strategy exploration for medical image classification.

Purpose of the Study:

• To propose a Deep Neural Network (DNN) edge-optimized Collaborative Inference strategy (MOCI) for medical image diagnosis.
• To address performance, energy, and communication constraints in mobile edge computing (MEC) environments for medical imaging.

Main Methods:

• MOCI combines compression techniques (coding, quantization) with Multi-Agent Reinforcement Learning (MARL) for dynamic DNN model segmentation and collaborative execution.
• Optimal transmission distance (Wasserstein) and Long Short-Term Memory (LSTM) networks are employed to enhance policy stability and adaptability to dynamic task complexity.

Main Results:

• The MOCI strategy significantly reduces processing latency (up to 38.5%) and energy consumption (up to 71%) for medical image diagnosis.
• Achieved these reductions with minimal impact on classification accuracy (less than 2% loss) compared to other inference strategies.

Conclusions:

• MOCI effectively solves collaborative inference tasks in medical image diagnosis within MEC environments.
• The strategy demonstrates wide potential applications in intelligent healthcare, promoting efficiency and quality of services.