Artificial intelligence (AI)-driven ensemble model for comprehensive chest X-ray abnormality detection and deployment

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiology and Medical Diagnostics

Background:

• Chest X-rays (CXR) are crucial for detecting thoracic abnormalities like tuberculosis (TB).
• A shortage of radiologists in peripheral areas hinders timely CXR interpretation.
• Existing screening methods face challenges in scalability and accessibility.

Purpose of the Study:

• To develop and validate DeepCXR v1.1, an AI tool for automated detection of chest X-ray abnormalities.
• To create a metadata-independent tool suitable for large-scale public health screening.
• To enhance early disease detection capabilities in resource-limited settings.

Main Methods:

• Trained an AI tool on over 282,000 annotated CXR images from 54,000 patients across 18 Indian centers.
• Employed a multi-model ensemble architecture with lung segmentation and lesion-specific models.
• Validated the tool on multiple datasets, including a prospective validation of 13,927 CXR images.

Main Results:

• Achieved a sensitivity of 92.2% and specificity of 77.4% in blind prospective validation.
• Demonstrated strong generalisability across diverse training and validation datasets.
• Validated by expert committees and health technology assessment panels.

Conclusions:

• DeepCXR v1.1 provides a scalable, interpretable, and robust AI solution for augmenting radiological screening.
• The tool can improve early detection of diseases through automated CXR analysis.
• Its offline functionality on basic hardware makes it ideal for resource-limited settings.