Predicting cognitive decline in Parkinson's disease using FDG-PET-based supervised learning

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Cognitive impairment is a common and progressive symptom in Parkinson's disease (PD).
• Predicting the progression from mild cognitive impairment (MCI) to dementia in PD (PDD) is a significant clinical challenge.
• Early and accurate prognosis is crucial for patient management and therapeutic strategies.

Purpose of the Study:

• To develop and validate a machine learning model for predicting dementia progression in PD patients with MCI.
• To utilize brain fluorodeoxyglucose-positron emission tomography (FDG-PET) scans for stratifying patients based on their risk of cognitive decline.
• To assess the model's performance in distinguishing between PDD converters and stable MCI patients.

Main Methods:

• A supervised learning algorithm, specifically a support vector machine (SVM), was trained on baseline FDG-PET scans of 43 PD-MCI patients.
• Patients were categorized based on progression to PDD or stable MCI over a 5-year period.
• The model's performance was evaluated using sensitivity, specificity, and Area Under the Curve (AUC) on independent validation datasets.

Main Results:

• The SVM model achieved 95% sensitivity and 91% specificity in distinguishing PDD converters from stable MCI patients in the training set.
• Independent validation demonstrated an AUC of 0.73, with 67% sensitivity and 80% specificity.
• The model identified characteristic patterns of brain hypometabolism and hypermetabolism and showed high sensitivity for PDD and dementia with Lewy bodies (DLB), but not for normal cognition PD.

Conclusions:

• FDG-PET imaging combined with an SVM classifier shows promise for accurately predicting dementia development in PD-MCI.
• This approach could serve as a valuable prognostic tool in clinical settings.
• The model's ability to differentiate various dementia subtypes warrants further investigation.