Tool development
In metastatic prostate cancer, metastases mostly localize to bone, and patients can have numbers of lesions varying from few to hundreds, and the best way to assess all of these lesions and quantify individual lesion response is with imaging. With a high number of lesions it is impractical to manually determine the individual response of lesions coming from sequential scans, thus evaluation of patients with metastatic prostate cancer often relies on PSA levels, pain indexing, and bone scan index. For this reason our group has come up with a Quantitative Total Bone Imaging (QTBI) methodology to automate the process and provide more quantitative information of individual lesion response and total patient response. In my graduate studies I developed a prototype of QTBI that has recently been developed into a clinical software approved by the FDA. Find out more at Response Heterogeneity in Metastatic Cancer.
Some of my work on QTBI improved the lesion localization step and full automation of the identification step. We developed a set of statistically optimized regional thresholds for detecting bone metastases in sodium fluoride (NaF) PET/CT images. In NaF PET scans there is uptake in osteoarthritis, or degenerative joint disease (DJD), as well as in metastatic lesions, which is hard to differentiate based solely on PET information. I developed a machine learning method for classifying lesions as benign or malignant using random forests. One of my current focuses is to develop similar methods for other PET modalities, such as PSMA PET or FDG PET.
Leukemia is a disease that impacts the reproduction of cells in the bone marrow and can be imaged using PET imaging. To assess the changes of bone marrow we have developed Quantitative Total Marrow Imaging (QTMI), for analyzing FLT PET images of cellular proliferation. QTMI is being tested in a large clinical trial to determine if we can predict early response of leukemia to chemotherapy.
