Automating Response Assessment for Cancer Immunotherapies on PET/CT
Immunotherapies are a new class of cancer therapies designed to stimulate a patient’s immune system to mount an anti-cancer response. While immunotherapies have improved survival times for patients with several metastatic cancers, challenges related to assessing patient response remain. Response rates are low and variable, immune-related adverse events caused by treatment are common, and the novel mechanisms of action of immunotherapy lead to novel patterns of response on imaging. We use molecular imaging with 18F-FDG and 18F-FLT PET/CT as a non-invasive and quantitative method to address these challenges in monitoring patient response to immunotherapy.
I am interested in borrowing concepts from computer vision in order to develop a comprehensive and consistent framework for conducting PET/CT-based immunotherapy response assessment. Problems I am interested in include organ segmentation, detection of abnormal patterns in images, and training classifiers for predicting treatment outcome. Ultimately, tools for the automated analysis of PET/CT images can be used to help make decisions to select, change, or stop treatment to optimize immunotherapy patient management.
Figure 1: Three metastatic melanoma patients undergoing immunotherapy monitoring with 18F-FDG PET/CT. PET and CT volume renderings are shown in greyscale with lesion-wise response maps overlaid. Patients A and B demonstrate mixed response, with some lesions disappearing or decreasing in FDG uptake, while other lesions appear or progress. Patient C demonstrates partial response to therapy, with most disease disappearing and one disease site responding. The high degree of heterogeneity between individual responses shown in these example patients motivate the development of new tools for assessing immunotherapy response.
Figure 2: A patient receiving combination immunotherapy (ipilimumab and nivolumab) for metastatic melanoma was diagnosed with immune-related colitis 206 days after baseline imaging. However, deep-learning-based segmentation of the bowel (purple contour) shows elevated bowel uptake on FDG PET images acquired prior to clinical diagnosis. The patient was also diagnosed with immune-related pneumonitis at day 284, which corresponds with the increased lung uptake measured on that scan.