Patients with metastatic prostate cancer typically develop bone lesions that populate throughout their whole body. The total number of metastatic lesions can range from a few to a few hundred. Studies have shown that it is not only important to investigate each lesion at a single time point (in one image), but it is also conducive to evaluate the functional disease changes across time (over multiple images). Such an evaluation gauges overall disease burden changes and highlights relative changes in each lesion, enabling physicians to assess the global effect of treatment and to target particularly adverse lesions with tailored therapies. Traditionally, getting to this point would present a physician with a heavy work burden involving the need to contour all lesions in each image. Such a task is infeasible for most clinics. As a solution to this problem, our group has developed automated FDA-approved tools to detect lesions in PET images for the purpose of disease evaluation. To accommodate a more robust patient and disease population, we are currently exploring artificial-intelligence (AI) based methods for lesion detection, classification, and segmentation.
As a first-year student, my research goals involve optimizing a Convolutional Neural Network (CNN) for the purpose of lesion segmentation in metastatic prostate cancer PET/CT images. This can be accomplished through modifying the images that get fed into the network (pre-processing), by altering the network itself, or by further evaluation of the outputted images (post-processing). With a dependable lesion segmentation tool, physicians can begin to quantitatively track global and local disease changes over time which may reaffirm the effectiveness of a treatment or may call for a more modified treatment approach. Moreover, automated segmentation techniques may lead to the development of more robust imaging biomarkers that may be used to model how patients will uniquely respond to treatment, which in turn provides the opportunity to preliminarily pose personalized treatment options.