Consistent Explainable features for Alzheimer’s Disease Assessment through Handwriting

Thomas Thebaud, Esther S. Oh, Laureano Moro-Velázquez, Najim DehakAlzheimer 2024 · DOI

TL;DR

This study leverages detailed analysis of handwriting captured on digital tablets to identify subtle motoric changes associated with Alzheimer's Disease (AD) and Mild Cognitive Impairment (MCI). By extracting hundreds of explainable features, researchers found that velocity and movement efficiency (measured by entropy) are key indicators, paving the way for a non-invasive, early diagnostic tool.

What Problem Does This Paper Address?

Early detection of neurodegenerative diseases like Alzheimer's Disease (AD) is crucial for timely intervention and management, but often challenging, especially in the initial stages. Motoric changes, including those affecting handwriting, are known to be among the earliest symptoms. However, quantifying these subtle changes in a reliable and objective manner has been difficult. This paper aims to address this by exploring a comprehensive set of explainable features derived from handwriting tasks performed on a digital tablet, seeking to identify consistent markers that differentiate individuals with AD/MCI from healthy controls.

Methods

The researchers collected handwriting data from 69 participants (25 with AD/MCI and 44 controls) performing 13 different tasks on a digital tablet. These tasks were categorized into four groups: Point tasks (e.g., holding the pen still), Spiral tasks (drawing spirals), Writing tasks (copying text), and Drawing tasks (drawing images). From the recorded pen movements (position, pressure, angle), they computed 367 "global explainable features." These features were then statistically analyzed using Welch's t-test with False Discovery Rate (FDR) correction to identify those that showed significant differences between the AD/MCI and control groups. The study also assessed the consistency of these significant features across different tasks within each category.

Datasets

The study utilized data from 69 participants, comprising 25 individuals diagnosed with Alzheimer's Disease (AD) or Mild Cognitive Impairment (MCI) and 44 healthy control participants. The data was collected using digital tablets during the execution of various handwriting and drawing tasks. No new benchmark datasets were introduced; rather, existing participants were categorized based on their clinical diagnosis.

Key Insights and Ideas

  • Handwriting as a Sensitive Biomarker: The study strongly reinforces the idea that handwriting, often overlooked, can be a highly sensitive indicator of early neurodegenerative changes.
  • Velocity and Efficiency are Key: The most significant differentiating features were not complex geometric patterns but rather fundamental aspects of movement: how fast participants moved their pen (velocity) and how efficiently they executed movements (measured by entropy, indicating randomness or unpredictability).
  • Task Consistency Matters: Identifying features that are consistently significant across multiple tasks within a group (e.g., all point tasks) increases their reliability as diagnostic markers.
  • Explainable AI in Healthcare: The focus on "explainable features" is crucial for clinical adoption, as it allows clinicians to understand why a particular assessment is being made, fostering trust and interpretability.

Key Results

Out of the 367 extracted features, 189 were found to be significantly different between the AD/MCI group and the control group after FDR correction (p < 0.05). Furthermore, 52 of these features were statistically significant across all tasks within at least one of the four task categories. The most impactful features were related to:

  • Velocity: Average speed, angular speed, and variations in speed during tablet and in-air movements.
  • Movement Efficiency: Measured by entropy, indicating less predictable or more "hesitant" movements in individuals with AD/MCI. For example, features like "Mean Speed in Air" and "Shannon entropy in Air" showed highly significant differences (p < 10^-5 and p < 10^-6 respectively) in tasks like "Copy Text" and "Copy Image."

Limitations and Future Work

The authors acknowledge that not all participants were able to complete all tasks, which could introduce some bias. Future work could focus on expanding the dataset to include a larger and more diverse population, exploring a wider range of handwriting tasks, and investigating the longitudinal progression of these handwriting markers over time. Further research into the specific neurological underpinnings of these observed motoric changes would also be beneficial.

Summary updated

Abstract

Abstract Background Motoric changes are one of the earlier changes in neurodegenerative diseases such as Alzheimer’s disease (AD), but they are often difficult to detect especially in the early stages. In the past, examination of handwriting has been used to investigate motoric changes as a proxy for cognitive impairment. In this study, we explore a wide array of explainable features from the handwriting of patients, obtained using signal processing, and their impact on assessment of AD and Mild cognitive impairment (MCI). Method We collected data from 69 participants, 25 with AD or MCI, and 44 for the control group (CTL) on a digital tablet. We analyzed 13 different tasks, categorized into 4 groups: Point tasks (maintaining the pen still, close to the tablet), Spiral tasks (drawing spirals), Writing tasks (paragraphs from different stimuli), and Drawing tasks (drawing various images). Note that not all the participants succeeded to finish all the tasks. From the drawings of each subject were computed multiple characteristics: derivates of the positions, pressions and angles, then we used multiple statistics to get 367 global explainable features. Those features were compared to obtain the ones providing consistently the most significant differences between the two studied groups using Welch's t‐test, with false discovery rate (FDR) correction. We checked their consistency by comparing which ones were significant across all the tasks of a certain group. Result Our results showed that out of the 367 features extracted, 189 features were significantly different between AD/MCI and the control group after FDR‐correction (p‐value&lt;0.05), and 52 were statistically significant for all the tasks of at least one group of tasks. Most of the significant features were directly related to the velocity with which participants executed the various tasks (average speed, angular speed, variations of the speed on tablet, etc.), and related to the non‐effective movements, measured by the entropies. Conclusion The analysis of a large variety of handwriting segments employing machine learning techniques can provide an assessment of patients with AD and MCI that may be useful in their early diagnosis.
BibTeX
@inproceedings{thebaud2024consistent,
  author = {Thomas Thebaud and Esther S. Oh and Laureano Moro-Velázquez and Najim Dehak},
  title = {{Consistent Explainable features for Alzheimer’s Disease Assessment through Handwriting}},
  booktitle = {Alzheimer},
  year = {2024},
  doi = {10.1002/alz.092906},
}
Figure