Intuitive method for improving the quality of Patlak parametric images of FDG and FLT

Objectives We developed a simple, yet robust, technique to improve tumor and anatomical visualization in Patlak images of FDG and FLT.

Methods For three patients, each undergoing separate FDG and FLT studies of head and neck cancer, dynamic PET/CT imaging was performed on a Siemens Biograph 16 scanner with an initial ten minute acquisition over the heart and a subsequent forty-five minute acquisition over the head and neck. Arterial input functions (AIFs) were derived from the initial cardiac frames and late-time venous samples and were used to calculate Patlak images. Once Patlak equilibrium was achieved, AIFs were accurately modeled as simple exponential decay and therefore voxel activities asymptotically increased as A(t)=A(t*)+a(1-exp(-b(t-t*))). The rate constant b reflects the late-time behavior of the input function and causes this asymptotic model (AM) to be very sensitive to noise at the voxel level. However, because b is constant across voxels, it was accurately determined from a large homogeneous region. Therefore, a linear asymptotic model, AM with region-determined b, was fitted to the time course of each voxel. Patlak images calculated from these processed dynamic sets (Patlak-P) were then compared to nominal Patlak and SUV images via qualitative observations and target-to-background (TB) ratios.

Results Visualization of anatomic structures and tumors in both FDG and FLT Patlak images was significantly improved by applying our technique. Boundary definition, particularly for small regions, was improved relative to SUV images. Patlak-P TB ratios of small (S) and large (L) regions were larger than SUV by 70% (S) and 50% (L) and Patlak by 103% (S) and 5% (L).

Conclusions We have demonstrated the feasibility of our technique from a retrospective study of dynamic FDG and FLT acquisitions. Our method is accurate and robust, because it linearizes a high fidelity model. Our first impressions are that this technique can be used in tumor delineation, detection of metastases, and cerebral metabolic studies