One of my interests is in extracting “features” from images— in as automated as a fashion as possible.  The goal is to move beyond simple “this tumor is big” to developing quantitative imaging phenotypes that can potentially have similar diagnostic and prognostic information as other methodologies such as traditional histology and genetics.  One of the most important aspects of MRI in particular is that it is perhaps the only sampling methodology available where the entire tumor can actually be visualized.  When working with any pathology specimen or genetic tissue– a small section is extracted and analyzed.  While this data is of course critical… variability of the tissue due to random selection of where exactly in the giant tumor a biopsy came from can potentially influence results and interpretation.

For example a “low” grade tumor in the brainstem has vastly different implications than if the same biopsy was taken from another region of the brain like the frontal cortex.  Treatment possibilities for a more accessible tumor (e.g. frontal lobe) vs an area that would be extremely difficult to operate on would thus produce vastly different outcomes… even if the genetics of the two samples were identical.