Predictive value of tumor uptake heterogeneity spatial patterns in sequential 18F-FDG PET for rectal cancer radio-chemotherapy response monitoring
Résumé
Objectives: 18F-FDG PET image-derived indices such as textural features have been recently introduced to provide characterization of uptake heterogeneity. The objective of this retrospective study was to assess their predictive value on sequential scans during treatment for rectal cancer. Methods: 28 patients were included; all underwent 3 scans, at baseline (PET1), 1 week (PET2), and 2 weeks (PET3) into treatment. Tumor response grade (TRG) was assessed by histopathology after surgery, and patients were classified as responders (TRG 1-2) and non-responders (TRG 3-5). Several local and regional uptake heterogeneity parameters (homogeneity, entropy) were extracted. The predictive value of these parameters at PET1, as well as their evolution at PET2 and PET3 was evaluated using receiver operating characteristics (ROC) area under the curves (AUC), and were compared to standard parameters (functional volume, SUV). Results: Prediction of no response using baseline PET1 values was not accurate (AUC values from 0.58 for homogeneity to 0.75 for SUVmean). Although results were not improved by considering differences with PET2 parameters, the use of PET3 derived values led to AUCs varying from 0.79 ({Delta}SUVmean, {Delta}TLG) to 0.88 for {Delta}homogeneity (94% sensitivity, 83% specificity). Combining 2 parameters (for example SUVmean and homogeneity) at PET1 or PET2 through logistic regression led to improved AUCs of 0.80 and 0.95 respectively. SUVmax consistently showed a lower predictive value. Conclusions: Characterization of the tumor uptake spatial heterogeneity may provide complementary information of added predictive value within the context of radio-chemotherapy response monitoring using sequential 18F-FDG PET scans in the management of rectal cancer. Combination of several parameters to derive more accurate predictive models is feasible however it will require training and validation in larger cohorts, which will be the focus of future studies.