From guideline to practice: A single-center evaluation of the ‘action corridor’ model and generative AI for PET/CT quality management

Abstract

Harmonizing institutional Standard Operating Procedures (SOPs) with international PET/CT guidelines is complex and resource intensive. This study evaluates the ‘Action Corridor’ model for structured ‘to-be’/‘as-is’ comparison and assesses a generative AI-assisted workflow for extracting quality-relevant parameters from heterogeneous documentation in a single-centre setting. Current EANM/SNMMI guidelines were systematically compared with institutional SOPs for [ 18 F]FDG, [ 68 Ga]Ga-PSMA, and [ 68 Ga]Ga-DOTATOC PET/CT using a four-category framework (Conformity, Specification, Justified Adaptation, Potential Inconsistency). A generative AI model extracted predefined quantitative parameters from unstructured SOPs. AI-extracted values were cross-checked against independent human review. Overall harmonisation was high across protocols. Quantitative deviations were mainly observed in uptake times (practice-to-guideline ratios 0.41–0.75), most pronounced for [ 68 Ga]Ga-DOTATOC (30 min vs. 55–90 min). Reduced effective radiation exposure was achieved for [ 18 F]FDG (ratio 0.83). No discrepancies were identified in the predefined quantitative parameters compared with human verification. The ‘Action Corridor’ model enables transparent identification of guideline–practice deviations while preserving justified local adaptations. Generative AI reliably supports structured parameter extraction in quality management workflows. In this single-centre proof-of-concept setting, the framework demonstrates methodological feasibility. Empirical scalability across institutions requires prospective multicentre validation. • The ‘Action Corridor’ model visualises guideline vs. practice deviations in PET/CT. • Generative AI effectively organises unstructured QM data for analysis. • Framework supports continuous improvement cycles (PDCA) required by ISO 9001.

Ähnliche Arbeiten