Clinical Trials

We are studying tumor heterogeneity and treatment resistance in numerous clinical trials with extensive imaging endpoints for different types of interventions, such as radiation therapy, chemotherapy and molecular targeted therapies (e.g., VEGFR TKI, AR-directed therapies).

Therapy that blocks the process by which tumors develop vascular structure
Assessment of the proliferative and vasculature “flare” during drug holiday using FLT PET/CT

Assess changes in tumor proliferation and vascular status using static and dynamic FLT PET imaging
Evaluate imaging biomarkers’ prognostic value for patient outcome (Responders vs. Non-responders
Future: Extend analysis to later in the treatment cycle (Cycle 3) to probe development of treatment resistance

Cancer often metastasizes (e.g., mCRPC), which are treated with multiple lines of therapies
Assessment of total disease burden and disease heterogeneity to assist in drug development and clinical decision making

Strong heterogeneity between responding and non-responding lesions, and dependence on the assessed imaging measure
Future: Evaluation of the emergence of resistance to multiple therapies

Dose painting: deliver non-uniform radiation dose to tumors according to spatially-varying radioresistance to increase tumor control probability
Response in canine sinonasal tumors using extensive molecular imaging

Bradshaw et al., J Nucl Med 2013 
Bowen et al., Radiother Oncol 2012
 Gregoire, Jeraj et al., Lancet Oncol 2012

Strong spatial heterogeneity of radioresistance based on using multi-modality multi-tracer PET imaging
Determine optimal dose painting method
Future: Dose painting in human patients

The current clinical standard for predicting response in AML (Day 14 bone marrow biopsy) has a very poor predictive value (NPV=64%)

Evaluation of early treatment response assessment using FLT PET/CT as a biomarker

Assessment of the total bone marrow through FLT PET/CT was shown to be excellent predictor of response
Multi-center clinical trial through ECOG-ACRIN in preparation
Future: Assessment of bone marrow and treatment response heterogeneity as a predictive biomarker

Radiomics is a technique to extract high-dimensional quantitative data from imaging ROIs
Machine learning classifiers can be trained to predict lesion outcomes from radiomics feature signatures
Research focus: predicting lesion-level response to immunotherapy in metastatic melanoma

Figure: workflow for lesion response prediction. Left: quantitative features describing lesion uptake, shape, and heterogeneity are extracted from patient imaging, and labelled with a response category. Middle: feature values, represented as vectors in a high dimensional feature space are paired with their corresponding label to serve as training data. Right: classification and regression models are trained on the labelled data to predict lesion response.

Lymphoma is a challenging disease to assess, as it impacts organs throughout the body such as lymph nodes, bone marrow, and spleen
It is thus a time-consuming process for physicians to diagnose and assess how lymphoma patients are responding to therapy
Our group is currently working on developing tools to automatically quantify and assess patients with lymphoma, for use in assisting physicians and providing a “second opinion” on diagnosis and response assessment
To develop and validate these tools, we are utilizing techniques such as convolutional neural networks, automatic segmentation methods, and hand-crafted feature extraction (radiomics)