Fleischner Society 2017 Pulmonary Nodule Follow-up: A Practical Guide

The Fleischner Society 2017 guidelines for incidental pulmonary nodules are the de facto standard for follow-up recommendations on chest CT. They are also the single most-cited reason for CMS quality measure compliance under MIPS Quality Measure 364. Despite that, most radiologists I know do not have the size and density thresholds memorized. Neither do I — and that is fine, as long as a tool applies them consistently.

This guide covers what the 2017 guidelines actually say, when they apply, and how to automate them in your reporting workflow.

When the 2017 Fleischner guidelines apply

The 2017 update (MacMahon et al., Radiology 2017) covers:

• Incidental pulmonary nodules detected on CT performed for any reason other than lung cancer screening
• Patients 35 years or older (younger patients have substantially lower malignancy risk)
• Solid, part-solid, or pure ground-glass nodules
• Single or multiple nodules

The guidelines do not apply to:

• Patients with known primary malignancy (different surveillance protocol — by oncology)
• Immunocompromised patients (different infection-vs-malignancy considerations)
• Lung cancer screening patients — use Lung-RADS instead
• Patients under 35 — different malignancy denominators
• Symptomatic patients with concerning clinical features (work up directly)

This is why an AI radiology assistant cannot blindly apply Fleischner to every chest CT. Modality, clinical context, and patient demographics all gate the recommendation. MyRadAgent uses this exact gating logic.

The size and risk thresholds, simplified

For solid nodules in low-risk patients (no smoking history, no other risk factors):

For solid nodules in high-risk patients (heavy smoking, family history, etc.):

For subsolid nodules:

The full table includes multiple-nodule logic, which most reporting tools struggle with — multiplicity changes the recommendation, especially for the dominant nodule.

What MIPS Quality Measure 364 actually requires

CMS evaluates MIPS 364 as the percentage of qualifying reports where Fleischner-based follow-up is documented appropriately. The compliant report needs:

1. A statement that the Fleischner guidelines were considered (citing the 2017 update is best practice)
2. A specific follow-up timeframe if a nodule is reported
3. A reason if no follow-up is recommended (e.g., < 6 mm in low-risk patient)

The most common compliance failure: reporting a 7 mm nodule, recommending "follow-up CT in 6 months," but failing to cite Fleischner. CMS may flag this as non-compliant even though the recommendation itself is correct.

How AI radiology tools should handle this

A good AI radiology assistant approaches Fleischner this way:

1. Detect any pulmonary nodule mentioned in findings (text NLP) or visible on uploaded images (vision NLP).
2. Extract size, density (solid / part-solid / pure GGN), and location.
3. Apply gating — is this lung cancer screening (use Lung-RADS instead)? Is the patient under 35? Known malignancy?
4. Look up the appropriate threshold for the patient's risk category.
5. Insert a guideline-citing recommendation: "Per Fleischner Society 2017 guidelines, follow-up CT chest in 6-12 months recommended."
6. Flag non-compliance if the radiologist's draft has a recommendation that conflicts with Fleischner.

MyRadAgent does all six steps automatically, with modality and clinical-context gating. The implementation is in mips_compliance.py — see also our guide on MIPS quality measures for radiology.

Common edge cases that trip up automation

Things that look like Fleischner cases but are not:

• "Pulmonary opacity" — Fleischner is specific to nodules, not opacities. AI tools that match on "opacity" alone over-flag.
• "Stable known nodule" — already documented, no new recommendation needed.
• "Pulmonary nodules consistent with metastases" — known malignancy excludes Fleischner.
• "Pulmonary nodules are not a concern on this study" — negation should suppress the recommendation.
• "Recommend thyroid US" appearing in the impression — completely unrelated; should not flip modality detection.

These are the exact failure modes that show up when automation is built on regex alone. A robust AI radiology tool needs proper negation handling and section-aware text scanning.

Recommendation

If your group reads more than 100 chest CTs per month, automating Fleischner application is one of the highest-ROI workflow changes available. Whether you build it in-house, use MyRadAgent, or use another radiology AI tool, the question is not whether to automate — it is which automation respects the modality, clinical context, and edge cases.

Try MyRadAgent free for 25 reports — see it apply Fleischner correctly across your real cases.

This guide is educational. It is not a substitute for the original Fleischner Society 2017 publication or for clinical judgment. Always consult the original article and your institutional protocols.