This question has been addressed by researchers at the University of Washington School of Medicine (Seattle, WA), who are developing a statistical model to predict quantitative changes in PET/CT due to respiratory gating. Such a tool could enable the use of patient-specific respiratory patterns to select between various motion compensation and motion suppression strategies prior to image acquisition (Phys. Med. Biol. 59 1027).

"The pattern of respiratory motion can tell us whether gating or other motion compensation methods have the potential to improve PET quantification," explained researcher Stephen Bowen. "As a simple counter-example, if the amplitude of motion is minimal (a few millimetres or less), then gating is unlikely to improve PET quantification. This avoids performing more complex data processing and analysis of respiratory-gated PET images if there will be no clinical benefit."

To develop their predictive model, the researchers used PET/CT data from 22 lung and liver cancer patients. They reconstructed static and gated PET images, and determined the image quality in terms of peak standardized uptake value (SUVpeak) for all lesions-of interest. They then calculated the relative difference in SUVpeak – %ΔSUVpeak – between PET images with and without gating for all patients. Gating was considered to be effective for a particular patient if %ΔSUVpeak was higher than 10%.

Respiratory features

The next step involved correlating a patient's %ΔSUVpeak with their respiratory trace features, measured during PET/CT scanning using Varian's real-time position management system. To do this, the researchers extracted eight respiratory features from each inhale-exhale cycle and tested these as independent variables for the prediction model. They observed that the 22 patients displayed a tremendous amount of inter- and intra-patient variability in their respiratory motion patterns.

As each respiratory trace contained hundreds of breathing cycles, Bowen and colleagues calculated the standard deviation (the variation from the average) and entropy (the uncertainty in a random variable) of each feature over the entire trace. This resulted in 16 features characterizing each respiratory trace.

Finally, the researchers constructed a prediction model linking quantitative imaging improvement (%ΔSUVpeak) with the features extracted from the respiratory motion traces. Using a step-wise feature selection approach, they determined the six most informative trace features and used these to construct the prediction model.

The researchers tested the model using a leave one-subject-out cross validation, in which the model is trained on respiratory traces of 21 patients and then used to predict the %ΔSUVpeak of the final patient (with the process repeated 22 times). The best prediction performances were achieved using an amplitude gating threshold of 30%, which gave a hit rate (the proportion of correct predictions of %ΔSUVpeak within a specified error limit) of 0.59 for a ±3% error, and 0.95 for an error of ±5% or ±7%.

The correlation coefficient between measured and predicted %ΔSUVpeak values was 0.88. When the prediction model was averaged over 22 cross-validation runs, the hit rates improved to 0.95 for a ±3% error, and 1 for a ±5% or ±7% error.

The authors note that these results indicate the feasibility of predicting changes in SUVpeak using only motion pattern features from respiratory traces. They propose a clinical workflow in which the patient's respiratory trace is acquired prior to imaging while they lie on the PET/CT scanner bed. The respiratory features are input into the prediction model to determine whether gating will increase SUVpeak by 10%, for example. This result will determine whether the patient is likely to benefit from respiratory gated PET/CT or will require more invasive motion management strategies.

Bowen and colleagues now plan to perform an independent validation of the prediction model on a second test cohort of patients. "We seek to apply these prediction models towards improved clinical utility and cost-effectiveness of PET/CT imaging and PET/CT-guided radiation therapy," he explained. "Specifically, respiratory pattern prediction models can flag potential challenges when using conventional PET/CT for thoracic or abdominal tumours. More broadly, the incorporation of cost-benefit metrics for various motion management techniques into these prediction models has tremendous potential to personalize patient care."

Related stories

• Harmonizing quantitative PET studies
• Quantitative PET: fundamental challenges
• Gating scheme cuts PET/CT mismatch
• Fully automated gating: a new era for PET?