Decomposing the Hounsfield unit: probabilistic segmentation of brain tissue in computed tomography

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Radiology

Background:

• Cranial computed tomography (CT) analysis is crucial for diagnosing neurological disorders.
• Accurate segmentation of brain tissues in CT images remains a challenge.
• Existing methods may lack precision in differentiating tissue types.

Purpose of the Study:

• To introduce and assess a standardized technique for brain segmentation in cranial CT scans.
• To utilize probabilistic partial volume tissue maps derived from high-resolution MRI data.
• To enable automated and precise tissue quantification in CT imaging.

Main Methods:

• Generated probabilistic tissue maps (white matter, gray matter, cerebrospinal fluid) from 600 normal brain MRIs.
• Registered MR images to MNI-152 standard space.
• Developed a customized CT reference image and applied inverse warping of probability maps for CT segmentation.
• Decomposed CT images into tissue-specific components.

Main Results:

• Demonstrated the feasibility of automated brain tissue segmentation in cranial CT scans.
• Highlighted the utility for unsupervised quantification and potential visual enhancement of CT images.
• Presented segmentation examples in pathological cases, including perfusion CT.

Conclusions:

• Automated cranial CT segmentation using MRI-derived tissue probability maps is achievable.
• Potential applications include quantifying white matter in leukoaraiosis, CSF in hydrocephalus, and gray matter in neurodegeneration.
• The technique supports perfusion map analysis with separate gray and white matter assessment.