5 - Performance Analysis of a Fully Automatic Real-Time Capable Pilot Tone Cardiac Triggering Framework

Background: The Pilot Tone navigator¹ (PT) is a novel approach for identification of both respiratory^2,8 and cardiac motion³ based on modulation of a high-frequency magnetic field by physiological motion. Modulations are detected using the MR scanner’s receiver system. We have previously shown that information contained in the PT signal can be retrospectively analyzed to obtain trigger information highly correlated to ECG ground truth³. Here we evaluate a fully automatic real-time (RT) processing framework prototype, designed for prospective triggering, in a patient cohort.

Methods: 18 patients were scanned at 1.5T (n=6) and 3T (n=12) using a free-running bSSFP sequence prototype (TR=2.5ms at 1.5T) and a free-running 3D radial⁶ prototype (TR=2.5ms at 3T) (MAGNETOM Sola & MAGNETOM Vida, Siemens Healthcare, Erlangen). PT data was collected in each readout, every 2.5ms. Measurements were performed during free-breathing (FB) and breathholds (BH). For testing the data were fed retrospectively to a prototype processing pipeline written in C++ running on a virtual environment simulating the scanner’s acquisition control system. During RT processing (filtering and trigger detection, Fig.1a), the multidimensional PT data need to be combined such that the cardiac signal is separated from respiration and other unwanted sources of motion³. This combination was calculated once by Independent Component Analysis (FastICA⁴) in each patient from 20s of FB data and applied to all subsequent scans. The cardiac motion component was chosen automatically based on power in the 0.8-2Hz frequency band.
Following trigger detection, PT triggers were matched to ECG and RR intervals determined. To evaluate overall performance in all patients, we used Bland-Altman (BA) analysis and give RR intervals as fractions of the average ECG RR interval in each scan.

Results: Figure 2a shows correlation of ECG vs. PT. BA analysis (Fig. 2b) shows the majority of RR intervals from PT well within the Limits of Agreement (LOA). This improves markedly when only evaluating triggers collected during breathholds (Fig. 2c). Out of 2015 evaluated ECG triggers, 274 could not be matched to a corresponding PT trigger, either due to failure of ECG or PT trigger detection. Out of 61 individual acquisitions, 4 had to be excluded because automatic ICA failed to detect the cardiac component. Three patients were excluded due to severe arrhythmia/suspected atrial fibrillation as the trigger algorithm currently is not designed for these cases.

Conclusion: We showed that fully automatic real-time processing of PT data can provide prospective triggers at both standard clinical field-strengths. We found trigger accuracy to be higher during BHs, likely due to imperfect separation of respiratory and cardiac motion during FB (see Fig.1b). The PT signal is directly related to cardiac motion and signal shapes are highly diverse among patients. More advanced triggering algorithms may be needed to account for this and ensure reliable triggering in all cases.