Five-class differential diagnostics of neurodegenerative diseases using random undersampling boosting.

Area of Science:

• Neuroscience
• Medical Imaging
• Biomarkers

Background:

• Accurate differentiation of neurodegenerative diseases is critical for treatment and clinical trials.
• Common neurodegenerative diseases include Alzheimer's disease, frontotemporal lobe degeneration, Dementia with Lewy bodies, and vascular dementia.
• Class imbalance in clinical datasets presents a challenge for developing effective diagnostic models.

Purpose of the Study:

• To develop a robust classification framework for distinguishing four common neurodegenerative diseases and subjective memory complaints.
• To address the challenge of imbalanced class prevalence in clinical datasets.
• To improve classification performance and identify important features and regions for diagnosis.

Main Methods:

• Extraction of multi-modal biomarkers, including imaging features (volume, region-wise grading) and non-imaging features (CSF measures).
• Implementation of the RUSBoost algorithm to handle class imbalance during classifier training.
• Integration of a multi-class feature selection method based on sparsity to enhance classification accuracy.

Main Results:

• The proposed framework achieved an overall accuracy of 75.2% and a balanced accuracy of 69.3% for five-class classification.
• The method significantly outperformed traditional classifiers like Support Vector Machine and Random Forest.
• Demonstrated the feasibility of the framework in supporting clinical decision-making for neurodegenerative diseases.

Conclusions:

• The developed classification framework effectively differentiates common neurodegenerative diseases using multi-modal biomarkers.
• The RUSBoost algorithm and sparsity-based feature selection are effective in handling class imbalance and improving diagnostic accuracy.
• This approach shows significant potential to aid clinicians in diagnosing neurodegenerative conditions and enriching clinical trials.