Bridging Brain and Machine: A Neurocognitive Framework for Early Alzheimer's Detection Through Hierarchical Speech...

Area of Science:

• Computational Neuroscience
• Artificial Intelligence in Healthcare
• Speech Signal Processing

Background:

• Alzheimer's disease (AD) impairs language networks, causing speech changes detectable early.
• Current automated speech analysis methods often miss subtle, diagnostically relevant temporal speech dynamics.

Purpose of the Study:

• To introduce the neuro residual acoustic learning (NeRAL) framework for early Alzheimer's disease detection using raw speech waveforms.
• To develop an end-to-end system inspired by biological auditory processing for enhanced speech analysis.

Main Methods:

• NeRAL utilizes acoustic-symbolic transformation, hierarchical temporal encoding with dilated convolutions, and attention mechanisms.
• The system processes raw audio waveforms, mimicking auditory cortex and prefrontal cortex functions.
• Evaluated on Mandarin (NCMMSC2021) and English (ADReSS) speech datasets.

Main Results:

• Achieved 86.55% accuracy on Mandarin extended utterances and 70.83% on English speech, outperforming existing methods.
• Demonstrated superior performance over multimodal and large language model approaches.
• Confusion matrix analysis identified specific misclassification patterns between AD and mild cognitive impairment, suggesting prodromal acoustic biomarkers.

Conclusions:

• The NeRAL framework offers a significant advancement in early Alzheimer's disease detection via speech analysis.
• Its biologically inspired architecture enhances the capture of subtle speech alterations.
• Results support the potential of acoustic biomarkers for non-invasive, early AD screening.