Quantitative evaluation of therapeutic response by FDG-PET-CT in metastatic breast cancer

Abstract

Purpose: To assess the therapeutic response for metastatic breast cancer with 18 F -FDG position emission tomography (PET), this retrospective study aims to compare the performance of six different metabolic metrics with PERCIST, PERCIST with optimal thresholds, and an image-based parametric approach. Methods: Thirty-six metastatic breast cancer patients underwent 128 PET scans and 123 lesions were identified. In a per-lesion and per-patient analysis, the performance of six metrics: maximum standardized uptake value (SUVmax), SUVpeak, standardized added metabolic activity (SAM), SUVmean, metabolic volume (MV), total lesion glycolysis (TLG), and a parametric approach (SULTAN) were determined and compared to the gold standard (defined by clinical assessment and biological and conventional imaging according RECIST 1.1). The evaluation was performed using PERCIST thresholds (for per-patient analysis only) and optimal thresholds (determined by the Youden criterion from the receiver operating characteristic curves). Results: In the per-lesion analysis, 210 pairs of lesion evolutions were studied. Using the optimal thresholds, SUVmax, SUVpeak, SUVmean, SAM, and TLG were significantly correlated with the gold standard. SUVmax, SUVpeak, and SUVmean reached the best sensitivity (91, 88, and 83%, respectively), specificity (93, 95, and 97%, respectively), and negative predictive value (NPV, 90, 88, and 83%, respectively). For the per--patient analysis, 79 pairs of PET were studied. The optimal thresholds compared to the PERCIST threshold did not improve performance for SUVmax, SUVpeak, and SUVmean. Only SUVmax, SUVpeak, SUVmean, and TLG were correlated with the gold standard. SULTAN also performed equally: 83% sensitivity, 88% specificity, and NPV 86%. Conclusion: This study showed that SUVmax and SUVpeak were the best parameters for PET evaluation of metastatic breast cancer lesions. Parametric imaging is helpful in evaluating serial studies.