Multimodal Analysis of Behavior During Stroop Test for Characterization of Alzheimer's Disease Signs

Trevor Meyer, Anna Favaro, Esther S. Oh, Pedro P. Irazoqui, Najim Dehak, Laureano Moro-VelázquezAlzheimer 2024 · DOI

TL;DR

This paper introduces a novel multimodal approach to analyze behavior during the Stroop Test, combining eye-tracking and speech analysis to detect subtle early signs of Alzheimer's Disease (AD). By quantifying deviations from healthy controls across numerous behavioral metrics, this method offers a more objective and detailed characterization of AD than traditional methods, showing promise for early detection and monitoring.

What Problem Does This Paper Address?

Assessing and characterizing Alzheimer's Disease (AD), especially in its early stages, is challenging. Current methods, like clinical rating scales and neuropsychological tests, often lack the granularity to capture subtle changes. These methods can be subjective, expensive, nonlinear, or rely on anecdotal evidence, making objective and consistent phenotyping difficult. The authors aim to address this by developing a more objective and detailed method to quantify behavioral changes associated with AD, particularly focusing on early-stage manifestations.

Methods

The researchers employed a multimodal analysis of participant behavior during the Stroop Test (ST). This involved recording both eye movements and spoken responses using an Eyelink Portable Duo system. They analyzed a comprehensive set of metrics, including blinking, saccades, fixations, gaze patterns, prosodic features of speech, accuracy, and timing. To quantify deviations from normal behavior, they normalized these metrics into positive z-scores based on the performance of a cognitively normal control (CTL) group. The overall behavioral abnormality was then summarized as the average z-score across 45 different metrics. A novel visualization, termed a "sign profile," was developed to represent these behavioral abnormalities.

Datasets

The study utilized data from two groups:

  • AD Subjects: 17 individuals diagnosed with Alzheimer's Disease. Eight of these subjects returned for 2-3 follow-up sessions, resulting in a total of 29 AD recording sessions.
  • Cognitively Normal Controls (CTL): 24 age-matched individuals.

The demographic descriptions of these groups are provided in Table 1, detailing total sessions, age, and MoCA scores.

Key Insights and Ideas

  • Beyond Single Scores: Traditional AD assessments often rely on single scores, which can miss nuanced changes. This multimodal approach captures a richer behavioral fingerprint, potentially revealing early signs that single-score tests might overlook.
  • Objective Quantification of Subjective Tasks: The Stroop Test, while a standard cognitive assessment, can be influenced by compensatory strategies. By analyzing eye movements and speech patterns, the researchers move beyond just the output to objectively quantify the process of task completion.
  • "Sign Profiles" for Intuitive Understanding: The development of a "sign profile" offers a visually intuitive way to represent complex behavioral data, making it easier for clinicians and researchers to understand a patient's specific pattern of AD-related changes.
  • Longitudinal Monitoring Potential: The inclusion of longitudinal data from some AD subjects suggests that this multimodal analysis could be valuable for tracking disease progression and the effectiveness of interventions over time.

Key Results

  • The average deviation across the analyzed behavioral metrics significantly correlated with MoCA scores (p < 0.001), indicating a strong link between objective behavioral changes and cognitive status.
  • The distributions of these behavioral metrics were significantly different between AD subjects and CTL subjects (p < 0.001).
  • The "sign profile" visualization of abnormality closely matched MoCA scores and clinician observations for all subjects.
  • Figure 1 demonstrates distinct "sign profiles" for a healthy control subject and for an AD subject across different visits, highlighting variations in their behavioral patterns.
  • Figure 2 shows correlations between specific metrics (e.g., Blink Vertical Coordinates, Duration Fixating per Word, Correct Utterance) and MoCA scores in AD subjects, with reported Spearman's correlation coefficients and p-values.

Limitations and Future Work

  • Sample Size: While the study included 17 AD subjects and 24 controls, the authors acknowledge the need for larger sample sizes to further validate their findings and generalize the results.
  • Generalizability: The study was conducted at Johns Hopkins University. Future work should aim to replicate these findings in diverse populations and clinical settings to ensure generalizability.
  • Early Stage Focus: The paper highlights the potential for detecting early clinical changes. Further research could focus on differentiating between very early, mild cognitive impairment (MCI) and early AD using this multimodal approach.
  • Integration with Other Modalities: While this study focuses on eye-tracking and speech, future work could explore integrating these with other physiological or neuroimaging data for an even more comprehensive understanding of AD.
  • Clinical Utility: Further validation is needed to establish the direct clinical utility of this method for routine diagnosis and patient management.

Summary updated

Abstract

Abstract Background Clinical rating scales and neuropsychological tests are commonly used for assessing sign and disease severity, yet lack detail in the early stages Alzheimer’s Disease (AD). Existing evaluation methods can be subjective, nonlinear, expensive, or reliant on anecdotal evidence making objective and consistent characterization and phenotyping of AD difficult. Multimodal analysis of patient behavior, rather than scoring of patient‐generated output which can be skewed by compensation strategies, presents a unique opportunity to objectively quantify AD related changes. Using the Stroop Test (ST), we present a multimodal analysis of eye movement and spoken responses during completion of ST tasks, with the goal to characterize patient behavior. Method 17 individual AD subjects and 24 age‐matched cognitively normal controls (CTL) were recruited from the Johns Hopkins Memory and Alzheimer’s Treatment Center and the Movement Disorders Clinic at the Johns Hopkins University School of Medicine. Eight subjects returned for two or three follow up sessions at least 20 weeks apart, totaling 29 AD recording sessions. An Eyelink Portable Duo in head‐free‐to‐move mode was used to record eye movements and spoken responses. We analyzed subjects’ interaction with ST using several metrics relating to blinking, saccade, fixation, gaze, prosodic, accuracy, and timing characteristics. We summarize patient behavior abnormality by scaling behavioral metrics into positive z‐scores using CTL performance and summarize the overall abnormality as the average z‐score across 45 metrics. Result The average deviation across metrics correlates with MoCA scores (p&lt;0.001). These metrics display significantly different distributions between the AD from CTL subjects (p&lt;0.001). A visualization we’ve coined as a “sign profile” summarizes patient behavior abnormality during ST. In all subjects our abnormality measurement closely matches MoCA scores and clinician observations. Conclusion AD subjects execute the ST with eye movement and speech behaviors which significantly differ from CTL subjects, and subjects with longitudinal recordings show encouraging results for the use of this analysis in monitoring. By utilizing measures from more than one feature, we detect early clinical changes a single‐score test might miss. This analysis offers comprehensive patient profiling, providing an intuitive and quantitative understanding of patient signs enabling more accurate detection and management of disease.
BibTeX
@inproceedings{meyer2024multimodal,
  author = {Trevor Meyer and Anna Favaro and Esther S. Oh and Pedro P. Irazoqui and Najim Dehak and Laureano Moro-Velázquez},
  title = {{Multimodal Analysis of Behavior During Stroop Test for Characterization of Alzheimer's Disease Signs}},
  booktitle = {Alzheimer},
  year = {2024},
  doi = {10.1002/alz.091604},
}
Figure