Novel quantitative generalized total effective entrapment (gTEE) metric for enhanced clinical outcome prediction: application to PET/CT imaging of pancreatic cancer

Objectives To investigate a novel quantitative metric, denoted generalized total effective entrapment (gTEE), derived by analogy to a model of generalized effective uniform dose (gEUD) as used in radiation therapy, for assessment of tumor burden.

Methods The metric is given by gTEE(a)=( Δv * Σ_iu_i^a )^1/a for a given tumor with N sub-volumes (voxels) of volume Δv each and PET uptake u_i (i=1…N). The metric is attractive as it generalizes commonly utilized metrics using a single additional degree-of-freedom or parameter ‘a’ (values of a=0, 1 and ∞ correspond to MTV, TLG and SUVmax, respectively). We applied this metric to n=45 pancreatic cancer patients with non-resectable locoregional disease imaged by FDG PET/CT scan at baseline. Kaplan-Meier survival analysis of overall survival (OS) was performed, where different threshold were considered and optimized for a given parameter ‘a’, and the hazard ratios (HR) between the higher and lower percentile groups were computed using Cox proportional hazards regression.

Results The proposed gTEE metric was optimized for a=10, i.e. when greater emphasis is put on PET uptake than volume, unlike TLG (a=1) which places equal emphasis on volume and uptake, and unlike MTV or SUVmax, which altogether neglect either uptake or volume, respectively. The optimized HR values for MTV, TLG, SUVmax and gTEE where 2.51, 2.51, 2.93 and 3.48, respectively. For the Cox model, the corresponding goodness of fit log-likelihood (LOGL) were -125.7, -125.7, -125.1 and -123.6, and the p-values (for the null hypothesis that HR=1) were 0.011, 0.011, 0.0077 and 0.0019. MTV&TLG (SUVmax) were outside the 95% (90%) confidence intervals of gTEE.

Conclusions The proposed framework enable an intuitive generalization of existing metrics, enabling placement of differing degrees of emphasis on tumor volume vs. uptake, to arrive at enhanced clinical outcome prediction.