Exploring physiological differences in time-frequency domain to improve tumor detectability for dynamic PET

The time activity curve (TAC) of FDG PET, which reflects glucose metabolic rate, often increases in cancer cell while decreases in normal tissue along with time, we studied this temporal difference in frequency domain to characterize the behaviors of the individual physiological function and TAC. An image enhancement algorithm was designed and tested based on this frequency domain property of TAC which showed a good performance. However, by simply applying the conventional frequency domain analysis, the distinguishable temporal characteristics appearing in TACs are somehow lost. In this paper, we further examined the TAC differences in time-frequency domain using wavelet based multi-channel analysis. By decomposing the TAC into different frequency bands and time slots, we found out that energy of TAC in tumor and background distributed differently in time-frequency domain. To validate these time-frequency differences of TAC, an energy ratio detection criterion is designed to distinguish malignance from normal tissue by evaluating the ratio of energies over two subsets of wavelet base functions in which malignant and normal TACs energy mainly lay. The phantom and clinical studies show that the tumor to background contrast can be improved by our method. These time-frequency properties can also be exploited for bases function selection for list mode based reconstruction method