Enabling real-time tuberculosis detection in hospital radiology through scalable actor-learner architectures for...

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Increasing demand for rapid tuberculosis detection in hospital radiology.
• Need for scalable and efficient automated screening solutions.
• Challenges in real-time analysis of high-throughput radiograph streams.

Purpose of the Study:

• To present a novel application of the IMPALA distributed deep reinforcement learning architecture for real-time chest radiograph analysis.
• To evaluate the impact of actor batch sizes and queue configurations on operational metrics.
• To enhance AI-driven infectious disease diagnostics in clinical settings.

Main Methods:

• Utilized the IMPALA distributed deep reinforcement learning architecture.
• Implemented a distributed actor-learner framework to decouple functionalities.
• Systematically evaluated modifications to actor batch sizes and queue configurations.
• Assessed operational metrics: throughput, processing latency, synchronization efficiency, and resource utilization.

Main Results:

• Optimized distributed setup significantly reduced response latency.
• Achieved substantial improvements in throughput.
• Consistently sustained high diagnostic accuracy (AUC-ROC).
• Demonstrated efficient handling of extensive radiograph streams.

Conclusions:

• The proposed approach automates triage and enhances prioritization for suspected tuberculosis cases.
• Supports clinical workflow scalability without compromising diagnostic accuracy.
• Represents an impactful advance in AI-driven infectious disease diagnostics for resource-sensitive settings.