Related Experiment Video
Updated: Jun 21, 2025
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
42.6K
D2BGAN: Dual Discriminator Bayesian Generative Adversarial Network for Deformable MR-Ultrasound Registration Applied
Mahdiyeh Rahmani1,2, Hadis Moghaddasi1,2, Ahmad Pour-Rashidi3
1Department of Medical Physics and Biomedical Engineering, Tehran University of Medical Sciences (TUMS), Tehran 1461884513, Iran.
Diagnostics (Basel, Switzerland)
|July 13, 2024
Summary
We developed a new AI method, Dual Discriminator Bayesian Generative Adversarial Network (D2BGAN), for accurate brain shift compensation in neurosurgery. This approach significantly improves the registration of intraoperative ultrasound (iUS) with pre-operative MRI scans.
Area of Science:
- Neurosurgery
- Medical Imaging
- Artificial Intelligence
Background:
- Neurosurgical navigation accuracy is compromised by brain shift.
- Intraoperative ultrasound (iUS) registration with pre-operative MRI is a common compensation strategy.
- Challenges include low ultrasound image quality and unpredictable brain deformation.
Purpose of the Study:
- To propose an automatic, unsupervised, end-to-end MR-iUS registration method named Dual Discriminator Bayesian Generative Adversarial Network (D2BGAN).
- To enhance the accuracy and robustness of neuro-navigation systems in the presence of brain shift.
Main Methods:
- Developed the D2BGAN, featuring two discriminators and a generator.
- Optimized the generator using a Bayesian loss function for improved functionality.
- Incorporated a mutual information loss function in the discriminator for similarity measurements.
Main Results:
- Achieved a mean target registration error (mTRE) of 0.75 ± 0.3 mm on RESECT and BITE datasets.
- Demonstrated an 85% improvement in mTRE compared to initial errors.
- The Bayesian loss function improved MR-iUS registration accuracy by 23% over typical loss functions.
Conclusions:
- D2BGAN significantly enhances MR-iUS registration accuracy for neurosurgical navigation.
- The proposed method effectively compensates for brain shift, preserving image intensity and anatomical information.
- This advancement offers improved precision and reliability in image-guided neurosurgery.
