A deep unrolled neural network for real-time MRI-guided brain intervention

Area of Science:

• Medical Imaging
• Neurosurgery
• Artificial Intelligence in Medicine

Background:

• Accurate navigation is crucial for neurological interventions like biopsies and deep brain stimulation.
• Real-time image guidance enhances surgical planning, with MRI being suitable for pre- and intra-operative imaging.
• A key challenge in real-time interventional MRI (i-MRI) is balancing spatial and temporal resolution.

Purpose of the Study:

• To develop a deep unrolled neural network, LSFP-Net, for real-time i-MRI reconstruction.
• To integrate LSFP-Net with an MR-compatible interventional device for a real-time MRI-guided brain intervention system.
• To evaluate the system's performance in phantom and cadaver studies.

Main Methods:

• Proposed LSFP-Net, a deep unrolled neural network for accelerated i-MRI reconstruction.
• Integrated LSFP-Net with a custom MR-compatible interventional device on a 3T MRI scanner.
• Evaluated system performance using phantom and cadaver studies.

Main Results:

• Achieved 2D/3D real-time i-MRI with temporal resolutions of 80/732.8 ms.
• Reported latencies of 0.4/3.66 s, including data communication, processing, and reconstruction.
• Obtained an in-plane spatial resolution of 1x1 mm².

Conclusions:

• The proposed LSFP-Net system enables real-time monitoring of remote-controlled brain interventions.
• The system demonstrates potential for seamless integration into diagnostic scanners for image-guided neurosurgery.
• This advancement offers improved precision and safety in neurosurgical procedures.