Standardized visual reading of F18-FDG-PET in patients with non-small cell lung cancer scheduled for preoperative thoracic lymph node staging

OBJECTIVES: Routine visual assessment of positron emission tomography (PET) for thoracic lymph node (LN) staging in patients with non-small cell lung cancer (NSCLC) is limited by a lack of reliable assessment criteria. This study evaluates the accuracy and inter-rater agreement of a standardized approach with unified windowing and a PET-based visual score.

MATERIALS AND METHODS: This retrospective analysis included pretherapeutic FDG-PET data of 86 patients with NSCLC. After standardized windowing (threshold: 2×liver SUVmean) the LN uptake was assessed visually by three independent readers with varying levels of experience using a 4-step score (1, LN uptake≤mediastinal blood pool structures (MBPS); 2, MBPS<LN<liver; 3, liver≤LN<'black'; 4, LN appears 'black'). ROC analyses and respective areas under the curve (AUC) based on histology/cytology as standard of reference. Agreement was analyzed with Cohen's kappa (κ, pairwise) and Fleiss' κ (overall). Subgroup analyses separated between hilar vs. mediastinal LNs, adenocarcinoma vs. squamous cell carcinoma and grading G1/2 vs. G3/4.

RESULTS: Fifty-four of the 278LNs (19.4%) were malignant (optimal cut-off to differentiate benign vs. malignant, score >3). The inexperienced (n=1), advanced (n=1), and expert readers (n=1) achieved similar accuracies of 93.5%, 91.4% and 92.1%, respectively (P>0.05 each). Cohen's κ ranged from 0.92 to 0.96 and Fleiss' κ was 0.93. ROC-analyses showed no significant differences between attendant readers within any subgroup (AUC, 0.92-0.96).

CONCLUSION: Applying unified windowing, the introduced PET-score achieved highly accurate and robust LN assessment. This approach may shorten learning curves of inexperienced readers, facilitate multicenter trials, and improve comparability of future studies.