Biomarkers

Area of Science:

• Neurology
• Computational Linguistics
• Artificial Intelligence

Background:

• Mild Cognitive Impairment (MCI) is a precursor to Alzheimer's disease, making early detection crucial for intervention.
• Linguistic markers in conversation offer a promising, cost-effective method for MCI identification.
• Analyzing the temporal dynamics of language, rather than aggregated data, may reveal subtle cognitive changes.

Purpose of the Study:

• To develop and evaluate a novel temporal harmonization method for enhancing MCI detection using linguistic markers.
• To investigate the effectiveness of analyzing fine-grained conversational sequences over aggregated data.
• To address challenges posed by individual speaking style variations in sequence models for cognitive assessment.

Main Methods:

• Utilized 6,771 conversations from 74 participants in the I-CONECT clinical trial.
• Extracted 99 linguistic features per conversation, including syntactic complexity and lexical diversity.
• Developed a temporal harmonization method using adversarial training (Seq2Seq, Subject Classifier, Cognitive Classifier) to mitigate subject-specific variations.

Main Results:

• Temporal sequences of language markers improved MCI detection compared to aggregated single-conversation outputs.
• The proposed temporal harmonization method significantly increased subject classification performance, achieving an AUC of 0.720 compared to 0.647 without harmonization.
• The approach demonstrated reasonable Area Under the Curve (AUC) performance using only semi-structured conversation features.

Conclusions:

• Temporal sequences of language markers offer significant benefits for detecting Mild Cognitive Impairment.
• Temporal harmonization effectively removes subject-specific linguistic variations, further boosting cognitive detection accuracy.
• This study highlights the potential of analyzing conversational dynamics for early MCI identification.