Pattern recognition for cache management in distributed medical imaging environments.

Area of Science:

• Computer Science
• Medical Informatics

Background:

• Traditional medical imaging repositories rely on costly indoor infrastructure.
• Cloud outsourcing offers advantages but faces challenges with communication latency due to large data volumes.
• Cache and prefetching mechanisms are crucial for mitigating latency but depend on accurate data selection.

Purpose of the Study:

• To develop a pattern recognition system using artificial neural networks with incremental learning.
• To evaluate the system's ability to predict user behavior and select data for cache and prefetching.
• To assess the accuracy of the pattern recognition model under various training conditions.

Main Methods:

• Implemented a pattern recognition system employing artificial neural networks with incremental learning.
• Trained and evaluated the model using both real-world and synthesized datasets.
• Assessed model accuracy with varying amounts of training data, including minimal initial samples.

Main Results:

• Incremental learning proved advantageous, achieving high accuracy even with minimal initial training data (1 week).
• The system demonstrated accuracy comparable to models trained with 75% of long-term data.
• Preliminary results showed a significant reduction in communication latency when the proposed solution fed a prefetching mechanism.

Conclusions:

• The developed approach shows promise for enhancing cache replacement and prefetching policies in medical imaging cloud environments.
• The system's effectiveness from initial deployment moments makes it a valuable tool for optimizing cloud-based medical imaging workflows.