Machine Learning for Predicting Individual Severity of Blepharospasm Using Diffusion Tensor Imaging.

Area of Science:

• Neuroimaging
• Neurology
• Medical Technology

Background:

• White matter abnormalities in blepharospasm (BSP) patients are indicated by diffusion tensor imaging (DTI) metrics like local diffusion homogeneity (LDH) and fractional anisotropy (FA).
• These DTI metrics correlate with BSP disease severity, but their ability to individually assess severity remains unclear.

Purpose of the Study:

• To investigate if a combination of machine learning and DTI metrics (LDH or FA) can accurately identify individual blepharospasm (BSP) severity.
• To differentiate between non-functionally and functionally limited BSP patient groups based on DTI data.

Main Methods:

• Forty-one BSP patients underwent DTI scans and assessment using the Jankovic Rating Scale.
• A machine learning model, employing beam search and support vector machines, utilized LDH or FA values from 68 white matter regions as input features.
• Patients were categorized into non-functionally and functionally limited groups based on their Jankovic Rating Scale scores.

Main Results:

• The machine learning scheme achieved high accuracy in classifying BSP severity: 88.67% with LDH and 85.19% with FA.
• Sensitivity for identifying functionally limited outcomes was 91.40% (LDH) and 85.87% (FA).
• Specificity for identifying non-functionally limited outcomes was 83.33% (LDH) and 83.67% (FA), with areas under the curve of 93.7% (LDH) and 91.3% (FA).

Conclusions:

• Combining local diffusion homogeneity (LDH) or fractional anisotropy (FA) measurements with advanced machine learning accurately identifies individual blepharospasm (BSP) disease severity.
• This DTI-based machine learning approach offers a reliable method for assessing BSP severity and patient functional limitation levels.