Clinical Scorecard: A Comprehensive Guide to the Role of Artificial Intelligence in Thoracic Imaging: Insights from the European Society of Thoracic Imaging (ESTI)

At a Glance

Key Highlights

• AI applications in thoracic imaging include noise reduction, lesion detection, segmentation, volumetry, and diagnostic support, but most lack sufficient clinical efficacy evidence.
• Computer-aided detection (CAD) tools enhanced by AI improve lung nodule detection sensitivity and reduce inter-observer variability, critical for lung cancer screening.
• Clinical acceptance of AI tools depends on demonstrated utility, workflow efficiency, cost-effectiveness, and clear medicolegal accountability.

Guideline-Based Recommendations

Diagnosis

• Use AI as an adjunct to radiologists for lesion detection and characterization, particularly in lung nodule detection.
• Recognize that current AI models require rigorous validation due to methodological flaws and biases in training data.
• Employ systematic approaches and adhere to published guidelines and checklists to ensure high-quality AI study design and evaluation.

Management

• Integrate AI tools to augment radiologist workflow, aiming to reduce tedious tasks and improve diagnostic accuracy.
• Consider hybrid radiology reports combining radiologist and AI-generated content with clear identification of AI contributions.
• Implement AI solutions that demonstrate workflow speed improvements or diagnostic enhancements without excessive cost.

Monitoring & Follow-up

• Maintain meticulous logs of AI decision history to ensure traceability and medicolegal accountability.
• Use blockchain-based electronic medical records where possible to track AI algorithm versions and contributions over time.

Risks

• Be aware of potential biases and systematic errors in AI models due to limited and non-representative training data.
• Recognize the risk of false positives that may slow workflow or reduce efficiency.
• Understand that lack of clinical validation may limit the usefulness of some AI applications.

Patient & Prescribing Data

Patients undergoing thoracic imaging, including those in lung cancer screening programs and with interstitial lung disease

AI tools improve detection sensitivity and reduce variability in lesion assessment but require careful integration to avoid workflow delays and ensure clinical utility.

Clinical Best Practices

• Use AI to assist, not replace, radiologists in thoracic imaging interpretation.
• Validate AI tools with multi-centric, well-curated datasets to minimize biases.
• Adopt hybrid reporting systems that clearly delineate AI and human contributions.
• Monitor AI performance continuously and document decision-making processes for accountability.
• Prioritize AI solutions that enhance workflow efficiency and are cost-effective.

References

• NHS AI dictionary
• European Society of Thoracic Imaging (ESTI) AI guidance
• Systematic review on AI in COVID-19 imaging
• CE-marked AI applications analysis
• Martini et al. study on CAD in lung nodule detection