Respiratory-induced liver motion prediction using ECG as surrogate signal

Area of Science:

• Medical Imaging
• Physiological Signal Processing
• Computational Modeling

Background:

• Liver motion during respiration complicates image-guided interventions.
• Accurate prediction of liver motion is crucial for precise needle insertion and treatment delivery.

Purpose of the Study:

• To investigate the electrocardiogram (ECG) as a noninvasive surrogate signal for modeling liver respiratory motion.
• To develop and validate a learning-based model for predicting liver motion using only ECG data.

Main Methods:

• A learning-based encoder-decoder model was trained to map ECG signals to liver motion.
• The model exclusively utilized ECG data, eliminating the need for additional imaging.
• Validation involved a human subject study with eight participants exhibiting diverse breathing patterns.

Main Results:

• The model achieved a mean absolute error of 2.83 mm during normal breathing and 4.02 mm across all patterns, with correlation coefficients > 0.90.
• Over 90% of predictions met acceptable error margins for needle insertion.
• Performance decreased with increased liver motion during deep breathing.

Conclusions:

• Electrocardiogram (ECG) is a viable noninvasive surrogate for predicting liver respiratory motion.
• This ECG-based approach is clinically accessible and enhances guidance for interventional procedures.