VQ-Wave: A Physics-Driven Spatiotemporal Deep Learning Approach for Noncontrast-Enhanced Lung Ventilation and...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonary Medicine

Background:

• Noncontrast-enhanced functional lung MRI is crucial for assessing lung physiology.
• Traditional spectral decomposition methods struggle with physiological nonstationarity and noise.
• Developing robust deep learning frameworks is essential for advancing lung MRI techniques.

Purpose of the Study:

• To develop a deep learning framework for noncontrast-enhanced functional lung MRI.
• To overcome limitations of spectral decomposition in the presence of physiological nonstationarity.
• To accurately estimate ventilation and perfusion parameters.

Main Methods:

• Introduced VQ-Wave (Ventilation/Q-perfusion Waveform-based Assessment of Variable Evolutions), a physics-driven spatiotemporal inception neural network.
• Trained the network on synthetic signal models simulating realistic nonstationary dynamics and noise.
• Validated performance against matrix pencil (MP) decomposition using numerical phantoms and in vivo functional lung MRI in healthy volunteers and cystic fibrosis patients.

Main Results:

• VQ-Wave demonstrated superior robustness to nonstationarity compared to MP decomposition in numerical benchmarks.
• In vivo, VQ-Wave accurately captured functional defects in cystic fibrosis patients, yielding stable ventilation and perfusion maps.
• Performance remained high even with scan times reduced from 45s to 15s, while MP decomposition degraded significantly.

Conclusions:

• VQ-Wave offers a robust, physics-driven neural network alternative to spectral decomposition for functional lung MRI.
• The framework effectively handles physiological irregularity and noise, enabling reliable imaging.
• VQ-Wave facilitates reliable functional lung imaging with substantially shortened acquisition protocols.