Fully-automated white matter hyperintensity detection with anatomical prior knowledge and without FLAIR

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Cerebral white matter hyperintensities (WMH) are common in elderly individuals and associated with cognitive decline.
• Accurate detection of WMH is crucial for diagnosing and monitoring neurological conditions.
• Existing methods often rely on fluid-attenuated inversion recovery (FLAIR) MRI sequences, which are not always available.

Purpose of the Study:

• To develop and validate a novel method for detecting WMH using standard T1-, T2-, and PD-weighted MRI sequences.
• To demonstrate the method's efficacy in elderly subjects, particularly in the absence of FLAIR images.
• To evaluate the performance of probabilistic models and Markov Random Field (MRF) frameworks for WMH detection.

Main Methods:

• Utilized run-time PD-, T1-, and T2-weighted structural MRI data with labeled training examples.
• Developed probabilistic models for WMH spatial distribution and neighborhood dependencies.
• Integrated intensity models within a Markov Random Field (MRF) framework for WMH inference.

Main Results:

• The method accurately detected WMHs in 114 elderly subjects from a dementia clinic.
• Standard MRF training and inference methods yielded robust results.
• Performance remained strong even when training and testing data originated from different scanners and subject pools.

Conclusions:

• The proposed method reliably detects WMH in elderly brains using conventional MRI sequences, offering an alternative when FLAIR is unavailable.
• Probabilistic modeling within an MRF framework is effective for WMH detection.
• The method demonstrates robustness and generalizability across different imaging datasets.