Cross modality generative learning framework for anatomical transitive Magnetic Resonance Imaging (MRI) from...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Magnetic Resonance Imaging (MRI) provides high-resolution anatomical detail but can be costly and time-consuming.
• Electrical Impedance Tomography (EIT) offers a potentially more accessible method for imaging, but typically yields lower resolution.
• Bridging the resolution gap between EIT and MRI is crucial for enhancing diagnostic capabilities.

Purpose of the Study:

• To develop and evaluate a cross-modality generative learning framework for synthesizing high-resolution MRI images from low-resolution EIT data.
• To investigate the efficacy of a cascaded Cycle Generative Adversarial Network (CycleGAN) model for this image conversion task.
• To assess the impact of multifrequency EIT inputs on the accuracy of the generated MRI-style anatomical references.

Main Methods:

• A cascaded CycleGAN model was designed, integrating EIT data collection, domain adaptation using MRI images, and cross-modality generation.
• Multifrequency EIT data (70 kHz, 140 kHz, 200 kHz) and T1-weighted wrist MRI images were acquired from 19 healthy volunteers.
• The proposed cascaded CycleGAN was trained and compared against end-to-end CycleGAN and Pix2Pix models using 713 paired EIT-MRI images.

Main Results:

• The cascaded CycleGAN achieved superior bone detection accuracy (0.97) compared to end-to-end CycleGAN (0.68) and Pix2Pix (0.70).
• Multifrequency EIT inputs reduced the normalized root mean squared error of the MRI-style anatomical reference from 67.9% ± 12.7% to 61.4% ± 8.8%.
• The framework successfully generated MRI-style anatomical references from EIT images with reduced bone-related errors and good accuracy.

Conclusions:

• The proposed cross-modality generative learning framework effectively synthesizes high-resolution MRI-style images from EIT data.
• The cascaded CycleGAN architecture demonstrates significant improvements in anatomical accuracy, particularly in bone structure delineation.
• This approach holds promise for enhancing the utility of EIT in clinical settings by providing detailed anatomical context typically seen in MRI.