Deep ensemble learning with transformer models for enhanced Alzheimer's disease detection.

Area of Science:

• Computational neuroscience
• Artificial intelligence in medicine
• Natural language processing for healthcare

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder with no cure.
• Early detection and intervention are crucial for managing AD progression.
• Current diagnostic methods require enhancement for improved accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for early Alzheimer's disease diagnosis using textual clinical data.
• To leverage data augmentation and an ensemble learning approach for improved diagnostic performance.
• To enhance feature extraction and text comprehension for more accurate AD detection.

Main Methods:

• A data augmentation technique was applied to textual data.
• A two-branch BERT-based deep learning model was proposed, incorporating BERT encoder, convolution, LSTM, and recurrent convolutional neural network layers.
• An ensemble learning approach fused the outputs of the two branches for a robust classification.
• The model was trained on clinical notes from AD patients and healthy controls.

Main Results:

• The proposed ensemble model achieved high performance on the Cookie Theft subset of the DementiaBank Pitt Corpus.
• Achieved 94.98% accuracy, 0.9523 F1-score, and 0.93 AUC.
• Demonstrated superior performance compared to existing state-of-the-art models for early AD diagnosis from text.

Conclusions:

• The developed BERT-based ensemble model shows significant promise for accurate and efficient early diagnosis of Alzheimer's disease.
• Textual data analysis using advanced deep learning techniques can effectively aid in AD detection.
• This approach offers a valuable tool for clinical decision-making and patient management in AD care.