Prediction of treatment response and recurrence with texture analysis or volumetric parameters on FDG PET in patients with esophageal cancer receiving neoadjuvant chemotherapy and surgery

Objectives To evaluate if texture features or volumetric parameters on FDG PET were useful in predicting treatment response of neoadjuvant chemotherapy and recurrence after subsequent surgery in patients with esophageal cancer, especially focusing on the texture features having no or weak associations with volumetric parameters.

Methods Pretreatment FDG PET/CT exams in 35 esophageal squamous cell carcinoma patients were analyzed retrospectively. Only tumors with metabolic tumor volume (MTV, SUV threshold 2.5) >10ml were included for accurate analysis. Texture features were obtained from a 3D ROI covering each esophageal lesion on PET. Texture features evaluated in this study were first-order and global features (SUV histogram), second-order and local ones (spatial gray level dependence matrices <SGLDM>), and higher-order and local ones (neighborhood gray-tone difference matrices <NGTDM>). MTV and total lesion glycolysis (TLG) were obtained as volumetric parameters in addition to SUVmax. Texture features and volumetric parameters were compared each other and to pathologic results of neoadjuvant chemotherapeutic response and recurrent status after subsequent surgery. Spearman rank correlation test and multivariate logistic regression analysis was performed.

Results MTV ranged from 12 to 191 (median: 34). Various Texture features correlated with MTV and TLG. Among texture features, SGLDM entropy exhibited an extremely strong positive correlation with MTV or TLG (rho >0.9). SGLDM uniformity, NGTDM contrast and complexity showed extremely negative correlations with MTV or TLG (rho <-0.9). Recurrence showed weak positive or negative correlations with various texture features. No association was observed between recurrence and volumetric parameters in this study. Multivariate logistic regression analysis revealed SGLDM entropy, inverse difference moment, and uniformity were the statistically significant factors to predict recurrence. Among them, only SGLDM inverse difference moment was the factor not associated with MTV, TLG, nor SUVmax. SGLDM entropy and uniformity were demonstrated to highly correlate with MTV, TLG, and SUVmax. Pathologic response showed no correlations with texture features, volumetric parameters, nor SUVmax.

Conclusions This study demonstrated that texture analysis on FDG PET was useful and better than MTV and TLG in predicting recurrence after neoadjuvant chemotherapy and surgery in esophageal cancer. However, it should be noted that various texture features had correlations with MTV, TLG, and SUVmax.