Clinical Report: Role of Artificial Intelligence in Thoracic Imaging

Overview

Artificial intelligence (AI) in thoracic imaging offers promising applications including improved image quality, automated lesion detection, and diagnostic support. Despite abundant evidence of AI accuracy, clinical efficacy data remain limited, with ongoing challenges related to validation and acceptance.

Background

AI mimics human intelligence to perform complex tasks, with machine learning (ML) and deep learning (DL) as key subfields enabling pattern recognition and feature extraction without explicit programming. In thoracic imaging, AI applications range from noise reduction to automated lesion segmentation and diagnosis support. However, many CE-marked AI tools lack robust scientific proof of efficacy, and methodological flaws have limited clinical utility, especially highlighted during the COVID-19 pandemic. Guidelines have been developed to improve study quality and facilitate clinical integration.

Data Highlights

An analysis of 100 CE-marked AI applications for clinical radiology found that 64% lacked scientific proof of efficacy. A systematic review of 62 ML studies for COVID-19 diagnosis and prognosis revealed all had methodological flaws, rendering none clinically useful. Studies show that CAD tools improve lung nodule detection sensitivity and reduce inter-observer variability, though they may increase reading times due to false positives.

Key Findings

• AI enhances thoracic imaging through noise reduction, lesion detection, segmentation, and diagnostic support.
• Most AI applications lack sufficient clinical validation despite high accuracy and precision in controlled settings.
• CAD tools improve lung nodule detection sensitivity and reduce variability but may slow reading times due to false positives.
• Acceptance of AI tools depends on demonstrated clinical utility, workflow efficiency, and cost-effectiveness.
• Hybrid radiology reports combining AI-generated content with radiologist oversight are anticipated to become standard practice.
• Traceability of AI contributions is essential for medicolegal accountability, potentially facilitated by blockchain-based records.

Clinical Implications

Clinicians should view AI as an adjunct to radiologists, aimed at improving diagnostic accuracy and efficiency while reducing tedious tasks. Adoption requires careful evaluation of clinical utility, cost, and integration into existing workflows. Maintaining radiologist oversight and ensuring traceability of AI contributions are critical for safe and responsible implementation.

Conclusion

AI holds significant potential to augment thoracic imaging but requires rigorous validation and thoughtful integration into clinical practice. Future developments should focus on improving clinical efficacy evidence, workflow integration, and medicolegal frameworks to realize AI's full benefits.

References

1. European Society of Thoracic Imaging (ESTI) -- A Comprehensive Guide to the Role of Artificial Intelligence in Thoracic Imaging