Data-driven respiratory gating for the uEXPLORER with fast dynamics

Objectives: Data-driven respiratory gating (DDG) allows the detection of motion without the use of hardware devices. Tracer dynamics, especially the fast dynamics immediately following tracer injection, may affect the performance of the conventional DDG approach. Research has shown that time-varying tracer distributions can cause errors in the estimated respiratory phase, in some cases even flipping the sign of the gating signal. The goal of this study is to reduce the influence of the tracer dynamics for data-driven gating methods for the uEXPLORER system. Due to the large axial FOV of the uEXPLORER system, it is not practical to use to the whole FOV for the extraction of respiratory signals. A previously developed DDG method was used. In this method, the regions with likely respiratory motions were first automatically identified. The center of mass of the regions in the listmode data was calculated and used as a surrogate for respiratory signal, followed by application of a standard band pass filter. FDG dynamic scans that lasted 300 seconds post-injection were used for validation. Three different approaches were used. In the first approach, the conventional method was directly applied. In the second approach, the data excluding the first 90 seconds were used to calculate the VOI, and the same VOI was applied on the whole dataset to generate the respiratory signal. In the third approach, additional sign corrections and amplitude corrections were applied to the results generated in the second approach. For the sign corrections, the distributions of the respiratory period were first calculated using the more stable period of the scan (90 seconds to 300 seconds), and a maximum-likelihood approach was used to determine the signs for each cycle during the first 90 seconds. For the amplitude correction, the low-frequency components of the amplitude (period larger than 20 seconds) were removed. Equal-count amplitude gating was used to generate gated sinograms, which were reconstructed to verify the proposed methods. With the first approach, no meaningful respiratory signal was extracted due to the influence of the tracer dynamics. In the second approach, even with the use of conventional band-pass filters for the extracted signal, an amplitude difference as large as a factor of 4 exists between the early phase and the later phase of the signal. With the use of the third approach, the difference of signal amplitude between the early phase and the later phase was greatly reduced. With the third approach, visible respiratory motion can be detected from the gated reconstructions using both the 300 seconds scan and early phase only (90 s scan), suggesting the new approach was able to extract the respiratory motion information even with fast-changing dynamics. We have developed a method that shows promising initial results for DDG with fast-changing dynamics using the uEXPLORER system. More studies in the future will be conducted to further validate these methods.