Program Information
Predicting Lung Tumor Regression: Combining FDG PET-Based Mechanistic Models and CT-Based Geometric Models
M Crispin Ortuzar*, P Zhang, J Jeong, A Rimner, J O Deasy, Memorial Sloan Kettering Cancer Center, New York, NY
Presentations
TU-H-CAMPUS-JT-5 (Tuesday, August 1, 2017) 4:30 PM - 5:30 PM Room: Joint Imaging-Therapy ePoster Theater
Purpose: To predict the volume regression of non-small-cell lung cancer tumors using a mechanistic radiobiological model with parameters derived from pre- and mid-treatment FDG PET images, as well as a CT-based model that was previously trained with a cohort of patient images. Ideally, the predicted residual tumor could serve as guide for non-uniform dose distributions.
Methods: Patients were treated with standard-fractionated radiotherapy, and received pre- and mid-treatment FDG PET/CT. GTVs were delineated for treatment planning. The prediction of the residual FDG-avid tumor was based on a novel radiobiological model of tumor control probability incorporating individual patient changes in the FDG PET uptake pattern during therapy. The simulation relies on the fraction of cells in each voxel that are proliferative or hypoxic. These parameters were obtained by comparing pre- and mid-FDG PET scans voxel-wise after co-registering the images. Two alternative predictions of the post-treatment volume were produced, with and without inter-voxel cell migration (vM and vNM, respectively). A published geometric model of volume regression was used to make an anatomical prediction of the post-treatment tumor volume (vG). Volume and spatial distributions were compared between the biological and geometric predictions in terms of overall and overlapping volume.
Results: Eight patients were analyzed. The predicted CT-based tumor volume reductions averaged 40%. In four cases, where the average initial FDG uptake was low, the FDG-based model predicted complete response, except in one case where vM=3 cc (100% overlap with vG). In the other four cases, vM was between 22-49 cc (72-100% overlap with vG), and vNM was 44-180 cc (with an overlap of 61-72%).
Conclusion: Our results suggest that FDG PET-based predictions of volume regression are relatively consistent with CT-based geometric model predictions, although a combined prediction could provide increased reliability. More patient datasets are needed to tune and validate the proposed model.
Funding Support, Disclosures, and Conflict of Interest: Research supported by Varian Medical Systems.
Contact Email: