The diagnostic utility of the lognormal behavior of PET standardized uptake values in tumors

A meta-analysis of data primarily from PET oncologic investigations using FDG PET was performed. Its purpose was to establish statistical features of the distributions of standardized uptake values (SUVs) as possible aids in the diagnostic process.

We obtained 1536 values of oncologic markers from patient studies of 40 investigations in the literature. Statistical parameters were tabulated for analysis.

A significant observation is that, unlike skewed SUV histograms, log10SUV has Gaussian behavior, which is not uncommon for biologic quantities. This was found for SUVs of FDG and 2 amino acids as well as a few other cancer markers. A possible model for explaining this is proposed. For FDG, the SD sigma of the log10SUVs for an average cancer category was 0.23. Examining data within the framework of the model points to physiologic factors as dominating SUV variability rather than PET protocols. When data for a single cancer category were available from multiple institutions, averages, mean(SUV)s, disagree beyond chance expectations. Diagnostic utility suggestions include a universal linear relationship between sensitivity and severity, defined as SUV/mean(SUV), on semilogarithmic probability paper; a generic receiver-operating-characteristic curve for all cancers; using [log10(mean(SUVmal)/mean(SUVnorm))] divided by (sigma(mal)2 + sigma(norm)2)(1/2) as a simple diagnostic effectiveness measure; and using Gaussian log10SUVs to avoid erroneous P values.

Using the logarithms of markers, such as SUVs, several advantages stemming from their Gaussian nature can be achieved with benefits ensuing to the diagnostic process.