Source-linked AI summary
Evaluation of handwriting kinematics and pressure for differential diagnosis of Parkinson's disease
Peter Drotár, Jiří Mekyska, Irena Rektorová, Lucia Masarová, Zdeněk Smékal, Marcos Faundez-Zanuy
TL;DR
Because PD diagnosis lacks an objective method and handwriting is disrupted by the disease, this study develops and evaluates a handwriting-based approach using the PaHaW database. It records eight tasks from 37 PD patients and 38 healthy controls, extracts kinematic and pressure features, and compares classifiers; the results support discrimination between the groups, while the authors call for broader and longitudinal validation.
Problem
PD diagnosis lacks an objective method, can take months, and has an approximately 25% probability of inaccuracy.
Method
The study introduces the PaHaW database and uses kinematic and pressure handwriting features with SVM, AdaBoost, and K-NN classifiers for binary PD-versus-control classification.
Results
82% classification accuracy was achieved using basic kinematic and pressure features, and both modalities contributed to discriminating PD patients from healthy subjects.
Takeaways & Limitations
Handwriting kinematic and pressure analysis can assist more accurate and objective PD diagnosis without replacing the clinician.
Takeaways & Limitations
The study included only PD patients and healthy controls, so discrimination from other diseases remains to be investigated.
Abstract
from arXiv · showhide
Objective: We present the PaHaW Parkinson's disease handwriting database, consisting of handwriting samples from Parkinson's disease (PD) patients and healthy controls. Our goal is to show that kinematic features and pressure features in handwriting can be used for the differential diagnosis of PD. Methods and Material: The database contains records from 37 PD patients and 38 healthy controls performing eight different handwriting tasks. The tasks include drawing an Archimedean spiral, repetitively writing orthographically simple syllables and words, and writing of a sentence. In addition to the conventional kinematic features related to the dynamics of handwriting, we investigated new pressure features based on the pressure exerted on the writing surface. To discriminate between PD patients and healthy subjects, three different classifiers were compared: K-nearest neighbors (K-NN), ensemble AdaBoost classifier, and support vector machines (SVM). Results: For predicting PD based on kinematic and pressure features of handwriting, the best performing model was SVM with classification accuracy of Pacc = 81.3% (sensitivity Psen = 87.4% and specificity of Pspe = 80.9%). When evaluated separately, pressure features proved to be relevant for PD diagnosis, yielding Pacc = 82.5% compared to Pacc = 75.4% using kinematic features. Conclusion: Experimental results showed that an analysis of kinematic and pressure features during handwriting can help assess subtle characteristics of handwriting and discriminate between PD patients and healthy controls.
MASAROVA´ b, Zdenˇek SME´KALa, Marcos FAUNDEZ-ZANUYc
The paper is a 2016 Artificial Intelligence in Medicine article on handwriting kinematics and pressure for differential diagnosis of Parkinson’s disease.
- The study is titled “Evaluation of handwriting kinematics and pressure for differential diagnosis of Parkinson's disease.”
- The authors are Peter Drotár, Jiří Mekyska, Irena Rektorová, Lucia Masarová, Zdeněk Smékal, and Marcos Faundez-Zanuy.
- The article appeared in Artificial Intelligence in Medicine, volume 67, in 2016, on pages 39–46.
1. Introduction
The introduction motivates handwriting analysis as a potential noninvasive aid for Parkinson’s disease, addressing difficult diagnosis through kinematic and pressure measurements and the PaHaW database.
- Parkinson’s disease is a common neurodegenerative movement disorder involving tremor, rigidity, bradykinesia, and impaired motor control.
- Current PD diagnosis lacks an objective method, can take months, and has an approximately 25% probability of inaccuracy.
- PD can disrupt practiced handwriting through slower, segmented movements, hesitations, and pauses.
- The paper extends prior work by introducing novel pressure features and the PaHaW database containing in-air/on-surface trajectories and pressure.
- The PaHaW database is intended to support predictive models for PD diagnosis and comparison of SVM, AdaBoost, and K-NN classifiers.
2. Parkinson’s disease handwriting (PaHaW) database
The PaHaW database contains standardized handwriting recordings from 37 patients with PD and 38 matched healthy controls performing eight tasks, with trajectories and pressure captured digitally.
- Subjects: The database includes 37 PD patients and 38 sex- and age-matched healthy controls.
- Data acquisition: The tablet recorded on-surface and in-air movements together with perpendicular pressure during each task.
- Data acquisition: Recorded time sequences contained x- and y-coordinates, timestamps, button status, pressure, and discrete time.
- Subjects: All participants used their dominant right hand, had at least ten years of education, and PD patients performed tasks under L-DOPA medication.
- Handwriting tasks: Participants completed eight handwriting tasks, including an Archimedean spiral and repeated cursive letters or syllables.
- Handwriting tasks: Additional tasks required writing simple Czech words continuously without lifting the pen and writing a longer sentence.
3. Methods and results
The study extracts kinematic and pressure features from tablet-recorded handwriting and evaluates SVM, AdaBoost, and K-NN classifiers for PD discrimination. Across tasks, performance varies, task fusion generally improves accuracy, and task 8 is the strongest individual task.
- Feature extraction: A stroke was defined as one continuous on-surface trace between successive pen lifts and used only to calculate stroke speed.
- Feature extraction: Handwriting features were computed from on-surface Cartesian movements and recorded pressure, including conventional kinematic and novel pressure measures.Pressure features included pressure-change counts, statistical functionals, and correlations between pressure and velocity or acceleration.
- Statistical classification: The SVM used an RBF kernel, with gamma γ controlling kernel width and C and γ optimized by grid search.
- Numerical results: Task 8 was the most discriminative individual task, while combining tasks noticeably improved classification accuracy.For individual tasks, pressure accuracy reached 74.2% for task 6 and 73.2% for task 8; feature fusion improved accuracy only for task 8.
4. Discussion
The study supports handwriting-based assistance for distinguishing Parkinson’s disease from healthy controls, while emphasizing that pressure and kinematic features capture complementary information. Interpretation remains limited by disease-group scope, task dependence, feature volume, and the need for longitudinal validation.
- Feature interpretation: Pressure features provide information not captured by kinematic features, whose exact relationships with Parkinson’s disease symptoms remain unknown.Kinematic features are influenced by tremor, muscle stiffness, rigidity, and movement-speed variation.
- Task design: Task selection strongly influences discrimination: sentence writing was most promising, while several tasks reached only 62%–66% accuracy.The sentence task may reveal Parkinsonian representations over a longer temporal extent, including micrographia-related size reduction.
- Clinical relevance: 82% classification accuracy was achieved using basic kinematic and pressure handwriting features, supporting their use for diagnostic assistance rather than clinician replacement.The approach is intended to provide more accurate and objective diagnosis assistance, and combining it with speech processing may improve prediction accuracy.
- Limitations: Nearly 200 features make clinical interpretation difficult, and reducing them to representative severity-related measures requires many subjects across symptom-severity levels.The authors identify a need to map a smaller feature set to standard metrics and quantify Parkinson’s disease severity.
- Limitations: The study included only Parkinson’s disease patients and healthy controls, so discrimination from other diseases remains untested.Future work should include cognitively characterized patients at different motor stages and groups such as progressive supranuclear palsy or Huntington’s disease.
- Limitations: The proposed handwriting biomarker is a first step requiring longitudinal and repeated-measurement studies because intentional and unintentional handwriting changes may affect classification.Standardized data collection across multiple occasions is identified as necessary for confirming conclusions and assessing early diagnosis or symptom monitoring.
Appendix A. Subject data
Appendix A provides a table of detailed clinical and demographic information about the study participants.
- Appendix A. Subject data: Appendix A presents detailed clinical and demographic information about participants.