Human-machine partnership with artificial intelligence for chest radiograph diagnosis

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Human-in-the-loop (HITL) AI offers a symbiotic approach, leveraging human and AI strengths.
• Current diagnostic methods face limitations that AI aims to overcome.

Purpose of the Study:

• Investigate a novel collective intelligence technology using swarm-based systems.
• Amplify diagnostic accuracy of networked human groups in real-time.
• Compare swarm technology against human experts and deep learning AI for pneumonia diagnosis.

Main Methods:

• Applied swarm-based collective intelligence technology to radiologists diagnosing pneumonia on chest radiographs.
• Compared the swarm technology's performance against individual radiologists.
• Benchmarked against two state-of-the-art deep learning AI models.

Main Results:

• Both swarm-based technology and deep learning AI achieved superior diagnostic accuracy compared to human experts alone.
• The combination of swarm-based technology and deep learning AI demonstrated superior performance over either method individually.
• The integrated HITL AI solution significantly enhanced diagnostic accuracy.

Conclusions:

• HITL AI, particularly swarm-based and deep learning combinations, significantly improves diagnostic accuracy in radiology.
• This approach offers a powerful tool for clinical AI deployment, enhancing patient care.
• Future clinical practice can benefit from integrating advanced AI solutions for improved diagnostic outcomes.