Aberrant Structure MRI in Parkinson's Disease and Comorbidity with Depression Based on Multinomial Tensor Regression...

Area of Science:

• Neuroimaging
• Radiology
• Machine Learning

Background:

• Depression is a common nonmotor symptom in Parkinson's disease (PD).
• The underlying neural mechanisms of PD with depression (DPD) are not fully understood.
• Current DPD diagnosis relies heavily on clinical assessments.

Purpose of the Study:

• To develop a novel imaging technique for diagnosing DPD.
• To differentiate between non-depressed PD (NDPD) patients, DPD patients, and healthy controls (HC).
• To identify specific brain microstructure differences in NDPD and DPD.

Main Methods:

• Utilized multinomial tensor regressions with whole-brain structural MRI data.
• Employed maximum likelihood estimation and gradient descent algorithms.
• Applied the method to classify 196 NDPD patients, 84 DPD patients, and 200 HC.

Main Results:

• Achieved high prediction accuracy (0.94) and multi-class AUC (0.98) for distinguishing between NDPD, DPD, and HC.
• Identified key discriminative brain regions, including cortical areas, cerebellum, brainstem, basal ganglia, thalamus, and limbic regions.
• Demonstrated the technique's ability to predict DPD development in PD patients.

Conclusions:

• The tensor regression imaging technique accurately diagnoses DPD without prior feature information.
• This method enhances understanding of brain abnormalities in DPD and PD.
• It is valuable for statistical analysis of complex MRI data in diagnosing depression comorbidity in PD.