Automated cortical thickness and skewness feature selection in bipolar disorder using a semi-supervised learning...

Area of Science:

• Neuroimaging
• Psychiatric Research
• Machine Learning in Medicine

Background:

• Bipolar disorder (BD) impacts brain structure, particularly areas crucial for emotion and cognition.
• Machine learning shows promise for automatically distinguishing BD patients from healthy individuals.

Purpose of the Study:

• To automatically identify brain regions most affected by bipolar disorder using neuroimaging data.
• To leverage machine learning for objective identification of disease-related brain alterations.

Main Methods:

• Utilized cortical thickness data from 58 regions of interest from MRI scans of 41 BD patients and 34 controls.
• Employed a semi-supervised machine learning approach to address diagnostic uncertainty.

Main Results:

• Achieved approximately 75% classification accuracy using mean thickness and skewness of up to five regions.
• Identified the parietal lobe and specific temporal sulcus areas as most involved in bipolar disorder.

Conclusions:

• Automatic selection of brain regions involved in BD is significant for understanding its pathogenesis.
• The identified regions, known to be affected in BD, can serve as potential markers of the disorder.