A Long Short-Term Memory Biomarker-Based Prediction Framework for Alzheimer's Disease

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Early prediction of Alzheimer's disease (AD) is crucial for patient management and clinical decision-making.
• Existing predictive methods require enhancement for improved accuracy and forecasting capabilities.
• Machine learning offers potential for developing advanced predictive models in neurodegenerative disease research.

Purpose of the Study:

• To develop and evaluate an automated machine learning (ML) based predictive system for early Alzheimer's disease (AD) detection.
• To forecast future biomarkers for patients with AD and mild cognitive impairment (MCI) over extended periods.
• To assess the efficacy of the proposed model against existing algorithms using a benchmark dataset.

Main Methods:

• A recurrent neural network (RNN) incorporating long short-term memory (LSTM) was employed to predict future biomarkers.
• Neuropsychological measures (NM) and magnetic resonance imaging (MRI) data were utilized as input features.
• Predicted biomarkers were processed through fully connected neural network layers for final AD/MCI classification.

Main Results:

• The developed ML model achieved a prediction accuracy of 88.24% on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• The model demonstrated superior performance compared to other available algorithms for AD prediction.
• The system successfully predicted patient biomarkers at multiple future time points (6, 12, 18, 24, and 36 months).

Conclusions:

• The proposed automated system shows significant promise for the early and accurate prediction of Alzheimer's disease.
• The LSTM-based RNN model effectively forecasts disease progression, offering valuable insights for clinical practice.
• This ML approach provides a robust tool for distinguishing between Alzheimer's disease and mild cognitive impairment, outperforming current methods.