I’m pleased to announce the publication of a new journal article resulting from my work in the Giger Lab at the University of Chicago. This article, “Effect of biopsy on the MRI radiomics classification of benign lesions and luminal A cancers” investigates radiomic feature distributions of benign lesions and luminal A cancers of the breast from a database involving lesions imaged EITHER pre-biopsy OR post-biopsy (it’s very hard to locate cases in which a person was imaged with MRI both before and after biopsy.) Our hypothesis was that some features will change as a result of the biopsy process or the presence of a biopsy clip. Because it is so hard to collect single cases in both conditions, we took an approach of comparing groups of lesions. It’s part of an overall effort by our lab to understand the effects of what goes into machine learning algorithms for the purposes of computer-aided diagnosis. A lesion could be the same as another one nominally but have experienced differences in its physical state (such as in biopsy) or in imaging protocol, and we want to better understand the possible impacts on computer-aided diagnosis.
As always, peer-review made a useful impact on the paper. One of the reviewers encouraged us to look at precision-recall curve performance in addition to the usual AUC metrics our lab uses, especially because the cancer prevalence was so different in our pre-biopsy and post-biopsy sets. So you’ll see that extension to the paper.
I always get a lot of questions from fellow conference goers about what it is like to work on research at a primarily undergraduate institution. It’s not easy with a full teaching load, 12 credit hours each semester! I’m very proud that this paper was produced almost entirely from conception to publication without any kind of teaching release. I worked hard but also was able to work with the extraordinary folks at the University of Chicago. It takes a collection of people to do good research!