Prospective Respiratory Motion Correction Using Machine Learning and Pilot Tone (PROMPT) in Cardiac MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular MRI

Background:

• Respiratory motion is a major challenge in cardiac MRI, causing image artifacts and reducing diagnostic accuracy.
• Existing prospective gating techniques have limitations in accurately tracking and compensating for respiratory motion.

Purpose of the Study:

• To implement and assess the feasibility of a Pilot Tone (PT)-based prospective gating and tracking technique called PROMPT.
• To utilize a long short-term memory (LSTM) neural network for predicting respiratory motion from PT signals.

Main Methods:

• A subject-specific calibration scan was performed to estimate respiratory motion from images and PT data.
• An LSTM model was trained to predict respiratory motion from PT signals.
• PROMPT used predicted slice-shifting parameters for real-time gating and slice position updates during respiratory-corrected scans, evaluated in healthy volunteers and patients undergoing late gadolinium enhancement (LGE).

Main Results:

• The LSTM model demonstrated high accuracy in predicting respiratory motion, outperforming linear and polynomial regression models (NRMSEdB of -7.20 dB, MAE of 1.97 mm).
• PROMPT significantly reduced residual in-plane motion during LGE imaging compared to free-breathing retrospective gating (1.22 ± 0.38 mm vs. 1.35 ± 0.48 mm).

Conclusions:

• The PROMPT technique, leveraging an LSTM-based predictive model, was successfully implemented for respiratory motion correction in cardiac MRI.
• PROMPT significantly reduced in-plane motion and showed potential for limiting through-plane motion in LGE imaging.