Radiology Report Quality Auditor

Radiology Report Quality Auditor is a professional QA skill for reviewing the internal consistency and quality of CT, MRI, X-ray, and mammography report text.

It is designed to identify reporting inconsistencies before finalization while preserving the radiologist's clinical authority. The skill audits the report as written: it does not establish a new diagnosis, invent findings, or replace professional image interpretation.

What it does

• Findings–Impression Consistency detects direct contradictions and incompletely reflected relevant findings.
• Laterality & Localization checks left/right consistency, anatomical location, segments, levels, thirds, and other spatial descriptors.
• Measurement QA validates linear dimensions, areas, volumes, units, and explicitly stated measurement axes.
• Morphology Tracking checks consistency of finding identity, count, shape, margins, and internal structure.
• CT-specific QA checks Hounsfield Unit (HU) values, attenuation terminology, and stated contrast-phase consistency.
• MRI-specific QA checks sequence-dependent signal descriptions, including T1, T2, FLAIR, STIR, DWI, ADC, and post-contrast statements when present.
• Technical Quality Review checks whether artifacts, partial coverage, or other stated technical limitations are appropriately reflected in the final conclusion.
• Longitudinal Comparison reviews documented change over time when prior report text is available and checks patient identity, modality, protocol comparability, finding identity, measurements, and stated dynamics.
• Confidence Language detects unexplained changes in diagnostic certainty between report sections using a bounded standardized confidence lexicon.
• Recommendation Consistency checks existing follow-up recommendations for finding identity, laterality, localization, interval, and consistency with documented change or technical limitations.
• Template / Copy-Paste Review identifies possible residual text from another patient, modality, body region, side, finding, or recommendation.

How it works

1. Submit a radiology report text for review, optionally including prior report text for comparison.
2. The skill identifies the imaging modality and confirms that it falls within the supported scope: CT, MRI, X-ray, or mammography.
3. It extracts and tracks report-supported findings and their anatomical, quantitative, morphological, and modality-specific attributes.
4. It compares those attributes across Findings, Impression, recommendations, and prior reports when available.
5. It produces a structured QA report with issues categorized as CRITICAL, MAJOR, MINOR, or INFO.

Longitudinal comparison

Direct formal longitudinal comparison is performed only between reports from the same imaging modality:

• CT ↔ CT
• MRI ↔ MRI
• X-ray ↔ X-ray
• Mammography ↔ Mammography

Reports from different modalities may be used as context but are not treated as directly interchangeable quantitative measurements.

If a prior report is mentioned but not provided, the skill does not independently confirm or reject the claimed change.

Ordinary longitudinal change is not automatically converted into RECIST response categories.

Frameworks & logic

The skill uses:

• a structured severity matrix;
• bounded terminology normalization;
• measurement-type and unit normalization;
• longitudinal-comparison decision rules;
• a standardized confidence-language framework;
• evidence-based issue tracking and deduplication.

No external medical API is required.

Use cases

• Auditing a draft CT, MRI, X-ray, or mammography report before final sign-off.
• Detecting Findings–Impression contradictions.
• Checking laterality, anatomical localization, and finding identity.
• Screening for measurement, unit, or axis inconsistencies.
• Reviewing documented longitudinal change.
• Checking whether technical limitations are reflected in the Impression.
• Checking the internal identity and laterality consistency of an already documented BI-RADS category without independently assigning or recalculating BI-RADS.
• Reviewing recommendation and follow-up interval consistency.

Safety and scope

This skill audits radiology report text only.

It does not:

• replace a radiologist;
• interpret raw DICOM images as a primary diagnostic task;
• establish a new diagnosis;
• invent findings;
• prescribe treatment;
• independently assign or recalculate RECIST, BI-RADS, LI-RADS, PI-RADS, Fleischner, TNM, or other specialized classification systems.

Known limitations

• Ultrasound, PET/CT, nuclear medicine, and interventional reports are outside the base scope.
• Primary image interpretation is outside the base scope.
• A prior report that is mentioned but not provided cannot be independently verified.
• CRITICAL findings require physician review.
• MAJOR findings should normally be reviewed before report finalization.