Structural analysis of solid tumors based on regularized tubular modeling

Intratumoral heterogeneity has been associated with treatment outcome in several cancer types, and statistical analyses of PET imaging data allow for its assessment non-invasively. Objective characterizations of heterogeneity have been found to correlate with other standard prognostic factors, for example in lung cancer, and are most commonly performed through texture analysis. The success of image-based tumor characterization relies on our capacity to derive analytical summaries that are biologically meaningful and therefore useful in making critical treatment-adaptive decisions. Here we consider a fully-automatic modeling of tumor development profiles derived at voxel level, from which we can assess intratumoral heterogeneity as well as other quantitative features. A detailed, non-robust initial tubular representation of the spatial uptake data is obtained from a crude input volume using a methodology presented recently. This automatically delineated tubular volume is then further adjusted by thin-plate smoothing splines, which allows us to control for subregions of the 3D volume with sparse sampling, and smoothed across via model regularization to yield 3D-coherent and robut tumor descriptors. The result is a localized assessment of intratumoral texture, development status and heterogeneity. We present a numerical analysis of the concept on simulated data as well as demonstration on clinical studies from two medical centers (located in the USA and Ireland). The features derived from this approach create an opportunity both for improved patient prognosis and for advanced image-guided treatment (e.g. by guiding biopsy or radiotherapy). This modeling technique can be applied to FDG uptake information or any other PET tracer, and is also applicable to PET/CT and PET/MR data. The concept also applies to a range of solid tumors.