Transcranial MR Imaging-Guided Focused Ultrasound Interventions Using Deep Learning Synthesized CT

Area of Science:

• Medical Imaging
• Neurosurgery
• Artificial Intelligence

Background:

• Transcranial MR imaging-guided focused ultrasound (MRgFUS) is a novel therapeutic technique for various disorders.
• MRgFUS treatment planning requires both CT for skull density and MR imaging for target identification.
• Current workflows for MRgFUS planning are complex and can be simplified.

Purpose of the Study:

• To investigate the feasibility of using deep learning to create synthetic CT (sCT) skull images from MR imaging.
• To assess the utility of these sCT images for MRgFUS treatment planning.

Main Methods:

• A U-Net neural network was trained and tested on MR imaging data from 41 subjects.
• The model's performance was evaluated using k-fold cross-validation.
• Acoustic properties and skull density were verified by comparing sCT with reference CT scans; simulations were performed.

Main Results:

• The deep learning model generated sCT images highly comparable to true CT scans (spatial correlation coefficient: 0.80 ± 0.08).
• Skull density estimation using sCT was reliable (r = 0.96).
• Simulations showed comparable peak temperatures and distributions between sCT and reference CT.

Conclusions:

• Deep learning techniques can generate accurate synthetic CT skull images from MR imaging.
• This method significantly simplifies the clinical workflow for transcranial MR imaging-guided focused ultrasound treatment planning.