Technical Note: Simultaneous segmentation and relaxometry for MRI through multitask learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Accurate segmentation and relaxometry of human brain tissues are crucial for neurological disorder diagnosis and monitoring.
• Current methods can be time-consuming and may lack simultaneous segmentation and relaxometry capabilities.

Purpose of the Study:

• To demonstrate a magnetic resonance (MR) signal multitask learning method for simultaneous 3D segmentation and relaxometry of human brain tissues.
• To develop a deep neural network capable of processing multicontrast brain images for enhanced analysis.

Main Methods:

• Utilized a 3D inversion-prepared balanced steady-state free precession sequence for in vivo multicontrast brain image acquisition.
• Trained a deep neural network with online-synthesized MR signal evolutions and labels, incorporating prior knowledge of T1 and T2 values for normal brain tissues.
• Validated the method on healthy volunteers, animal models, and a prostate patient.

Main Results:

• The method achieved rapid whole-brain segmentation and relaxometry in approximately 5 seconds for healthy volunteers.
• Estimated apparent T1 and T2 maps correlated well with known brain anatomy and were unaffected by coil sensitivity variations.
• Successfully segmented gray matter, white matter, and cerebrospinal fluid (CSF), and generated synthetic T1- and T2-weighted images.

Conclusions:

• The proposed multitask learning method enables direct generation of brain apparent T1 and T2 maps and synthetic weighted images.
• This approach facilitates simultaneous segmentation and relaxometry, offering a significant advancement in brain tissue analysis.