Optimized sequential classification models for mild cognitive impairment screening based on handwriting and speech...

Area of Science:

• Neuroscience
• Computational Linguistics
• Medical Informatics

Background:

• Handwriting and speech analysis are key biomarkers for detecting cognitive decline, crucial for early diagnosis of Alzheimer's disease (AD) and mild cognitive impairment (MCI).
• Existing unimodal diagnostic approaches for AD and MCI face limitations in classification accuracy, potentially missing the combined diagnostic power of handwriting and speech data.

Purpose of the Study:

• To develop and evaluate an innovative multimodal classification model integrating handwriting and speech analysis for enhanced detection of MCI.
• To overcome the limitations of single-modality approaches by leveraging synergistic data fusion for improved diagnostic accuracy.

Main Methods:

• A multimodal classification model utilizing gated recurrent units (GRU) and an attention mechanism was developed, processing handwriting and speech data as sequential inputs.
• The model was validated on a dataset of 41 participants (20 MCI, 21 cognitively normal) using 10-fold cross-validation to ensure robustness against small sample size.

Main Results:

• The multimodal model achieved a classification accuracy of 95.2% for distinguishing MCI from cognitively normal individuals.
• This performance represents a significant improvement over single-modality approaches, demonstrating the effectiveness of cross-modal fusion for enhanced classification.

Conclusions:

• The proposed GRU-based model effectively fuses handwriting and speech data, significantly improving early MCI detection compared to unimodal methods.
• This approach shows promise for primary healthcare settings and provides a foundation for future research, including classification of MCI and AD stages.