Predicting Alzheimer's Disease Progression Using a Versatile Sequence-Length-Adaptive Encoder-Decoder LSTM...

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Informatics

Background:

• Early and accurate Alzheimer's disease (AD) detection is crucial for effective treatment and management.
• Predicting disease progression using existing diagnostic scores and clinical status at single time points remains challenging.
• Current deep learning models for AD detection often lack stability due to insufficient adaptive training on longitudinal data.

Purpose of the Study:

• To develop an adaptive deep learning model for predicting individual Alzheimer's disease diagnostic status over a six-year period.
• To enhance prediction performance by incorporating longitudinal patient data, improving with each additional patient visit.
• To address the limitations of non-adaptive deep learning models in real-world clinical settings.

Main Methods:

• Development of a Sequence-Length Adaptive Encoder-Decoder Long Short-Term Memory (SLA-ED LSTM) model.
• Utilizing longitudinal data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) archive.
• Dynamically adjusting the decoder LSTM to accommodate variable training and inference sequence lengths.

Main Results:

• The SLA-ED LSTM model demonstrated high predictive accuracy.
• For an inference length of one, a sequence length of nine visits yielded the highest average test accuracy (0.920) and AUC (0.982).
• The model significantly outperformed state-of-the-art methods in predicting disease progression.

Conclusions:

• Longitudinal data from approximately nine patient visits are sufficient to capture meaningful cognitive changes for accurate AD progression prediction.
• The proposed SLA-ED LSTM model offers a stable and improved approach for early Alzheimer's disease detection and progression forecasting.
• This adaptive deep learning strategy holds promise for enhancing clinical decision-making in Alzheimer's disease management.