Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detecting Neurodegenerative Diseases using Frame-Level Handwriting Embeddings

Sarah Laouedj, Yuzhe Wang, Jesus Villalba, Thomas Thebaud, Laureano Moro-Velazquez, Najim Dehak

TL;DR

The paper investigates handwriting-based detection of neurodegenerative diseases by representing signals as frame-based, multi-channel spectrograms and evaluating CNN and CNN-BLSTM classifiers. It contrasts fixed-size versus frame-based preprocessing and analyzes various signal channels and handwriting tasks to identify robust ND signatures across AD, PD, PDM, and CTL. Key findings show high AD vs CTL discrimination (F1 up to 89.8% on Point tasks), variable window-length effects across conditions, and that velocity and pressure cues are particularly informative, with CNN generally outperforming CNN-BLSTM. The approach offers a non-invasive, scalable avenue for early ND screening and highlights task- and channel-specific strategies to improve differential diagnosis, with potential for multimodal extension.

Abstract

In this study, we explored the use of spectrograms to represent handwriting signals for assessing neurodegenerative diseases, including 42 healthy controls (CTL), 35 subjects with Parkinson's Disease (PD), 21 with Alzheimer's Disease (AD), and 15 with Parkinson's Disease Mimics (PDM). We applied CNN and CNN-BLSTM models for binary classification using both multi-channel fixed-size and frame-based spectrograms. Our results showed that handwriting tasks and spectrogram channel combinations significantly impacted classification performance. The highest F1-score (89.8%) was achieved for AD vs. CTL, while PD vs. CTL reached 74.5%, and PD vs. PDM scored 77.97%. CNN consistently outperformed CNN-BLSTM. Different sliding window lengths were tested for constructing frame-based spectrograms. A 1-second window worked best for AD, longer windows improved PD classification, and window length had little effect on PD vs. PDM.

Detecting Neurodegenerative Diseases using Frame-Level Handwriting Embeddings

TL;DR

The paper investigates handwriting-based detection of neurodegenerative diseases by representing signals as frame-based, multi-channel spectrograms and evaluating CNN and CNN-BLSTM classifiers. It contrasts fixed-size versus frame-based preprocessing and analyzes various signal channels and handwriting tasks to identify robust ND signatures across AD, PD, PDM, and CTL. Key findings show high AD vs CTL discrimination (F1 up to 89.8% on Point tasks), variable window-length effects across conditions, and that velocity and pressure cues are particularly informative, with CNN generally outperforming CNN-BLSTM. The approach offers a non-invasive, scalable avenue for early ND screening and highlights task- and channel-specific strategies to improve differential diagnosis, with potential for multimodal extension.

Abstract

In this study, we explored the use of spectrograms to represent handwriting signals for assessing neurodegenerative diseases, including 42 healthy controls (CTL), 35 subjects with Parkinson's Disease (PD), 21 with Alzheimer's Disease (AD), and 15 with Parkinson's Disease Mimics (PDM). We applied CNN and CNN-BLSTM models for binary classification using both multi-channel fixed-size and frame-based spectrograms. Our results showed that handwriting tasks and spectrogram channel combinations significantly impacted classification performance. The highest F1-score (89.8%) was achieved for AD vs. CTL, while PD vs. CTL reached 74.5%, and PD vs. PDM scored 77.97%. CNN consistently outperformed CNN-BLSTM. Different sliding window lengths were tested for constructing frame-based spectrograms. A 1-second window worked best for AD, longer windows improved PD classification, and window length had little effect on PD vs. PDM.

Paper Structure

This paper contains 16 sections, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Dataset
  4. Tasks
  5. Point Tasks
  6. Spiral Tasks
  7. Writing Tasks
  8. Drawing Tasks
  9. Preprocessing
  10. Architectures
  11. Experiments Setup
  12. Results and Discussion
  13. Frame-Based Spectrograms with CNN and CNN-BLSTM Models
  14. Channels Combination Evaluation with Fixed-Size Spectrograms
  15. Task-Specific Classification Using Optimal Channel Combinations
  16. ...and 1 more sections

Figures (2)

  • Figure 1: Overview of the architectures used in the experiments: The top row represents the CNN architecture with Fixed-Size Spectrograms as input. The middle and bottom rows represent the CNN and CNN-BLSTM architectures using Frame-Based Spectrograms. In the Frame-Based approach, spectrograms were decomposed into smaller frames with a sliding window, and the extracted features were averaged across frames before passing through fully connected layers for classification.
  • Figure 2: F1-score performance of CNN and CNN-BLSTM models using Frame-Based Spectrograms with different window lengths on AD vs CTL, PD vs CTL, and PD vs PDM. The solid lines represent results from the CNN architecture. The dashed lines represent results from the CNN-BLSTM architecture. The window lengths varied from 25ms to 1.5s, with F1-scores plotted for each experiment.