Intra-tumor heterogeneity on baseline ¹⁸F-FDG PET images characterized by textural features predicts response to concomitant radio-chemotherapy in esophageal cancer

Objectives [18F] 2-Fluoro-2-deoxy-D-glucose (FDG) PET imaging is commonly used in response to therapy and prediction studies. The Standardized Uptake Value (SUV) is the most frequently employed quantitative index within this context. In this work we investigate the potential value of new indices assessing tumor heterogeneity on baseline FDG PET scans of patients with esophageal cancer for assessing response to concomitant radio-chemotherapy.

Methods 41 patients with a newly diagnosed esophageal cancer treated with combined radio-chemotherapy were included in this retrospective study. Patients were classified as complete (CR), partial (PR) or non-responder (NR) according to the RECIST criteria based on the use of follow up CT examinations. An automatic and reproducible algorithm was used for tumor volume delineations on the baseline scans. On these defined functional tumor volumes four heterogeneity parameters (two for tumor heterogeneity quantification at a local level and two at a regional level) computed by a textural analysis were extracted. These parameters were compared to more traditional PET image derived indices (SUV_mean, SUV_max, SUV_peak) for assessing therapy response using the Kruskal-Wallis test and ROC analysis.

Results The parameters describing both the local and regional tumor heterogeneity were able, in contrast to any of the SUV measurements, to significantly differentiate all three patient groups (p<0.0006 and p=0.0002 for the local and regional parameters respectively). ROC analysis showed that tumor textural features can provide NR, PR and CR patient identification with higher sensitivity (76%-92%) than any SUV measurement (46%-62%).

Conclusions Tumors heterogeneity quantification performed with textural features extracted from baseline ¹⁸F-FDG PET images allow for the best stratification of esophageal carcinoma patients in the context of therapy response prediction