2026 IASLC CT Screening and Early Detection Symposium-WS02.16 - Talk 2: AI for Lung Cancer Detection and Exclusion

Pdf Summary

The document reviews how AI is being used in lung cancer screening and nodule assessment, with a focus on two goals: improving cancer detection and safely ruling out benign disease to reduce workload.

Key points include:

- Lung nodules are very common in screening CT scans of people at high risk for lung cancer, but only a minority are malignant.
- AI can help with <strong>longitudinal nodule matching</strong> across follow-up scans, reducing manual tracking work, although performance drops when many nodules are present.
- A <strong>systematic review and meta-analysis</strong> of externally validated AI models for malignancy classification found strong overall performance: pooled sensitivity 88%, specificity 75%, and AUC 0.89.
- In external validation studies, deep learning models trained in the US maintained good discrimination in European screening cohorts.
- In the <strong>LUNA25 challenge</strong>, the best AI system outperformed the average of 65 radiologists for classifying indeterminate 5–15 mm nodules.
- Beyond detection, several studies show AI can act as a <strong>first reader</strong> to rule out negative baseline scans, potentially reducing unnecessary follow-up and radiologist workload while maintaining very high sensitivity.
- For small nodules (5–15 mm), one multicenter validation found an AUC of 94.5% and an NPV of 99.5%, suggesting that about 1 in 5 participants might be safely ruled out.
- Other screening datasets showed AI could have fewer missed referrals than radiologists and substantially reduce reading burden.

The overall message is that AI is promising for lung cancer screening, especially for nodule classification and benign rule-out, but prospective validation and clinically defined thresholds are essential before routine use.

Asset Subtitle

Marjolein Heuvelmans

Meta Tag

Concept Pulmonary Nodule

Concept Artificial Intelligence

Concept Malignancy Classification

Concept Malignancy Assessment

Concept Prospective Validation

Speaker Marjolein Heuvelmans

Keywords

lung cancer screening

pulmonary nodules

artificial intelligence

nodule classification

longitudinal nodule matching

systematic review

meta-analysis

deep learning

radiology workload reduction

benign disease rule-out

Pulmonary Nodule

Artificial Intelligence

Malignancy Classification

Malignancy Assessment

Prospective Validation

Marjolein Heuvelmans