Ameur Bensefia

Parkinson’s disease (PD) affects fine motor control and produces measurable abnormalities in handwriting and drawing. This study proposes a rigorously evaluated multimodal framework for PD detection that combines a Vision Transformer (ViT) for spiral and meander image analysis with an XGBoost classifier operating on 54 carefully engineered kinematic features extracted from multichannel handwriting signals. To assess how multimodal integration should be performed, both intermediate feature-level fusion and late decision-level fusion were evaluated under a strict 5-fold subject-wise cross-validation protocol, with supplementary sample-level analysis, on the NewHandPD dataset. The visual stream consistently outperformed the acquisition stream as a single modality, while both fusion strategies improved performance by exploiting complementary spatial (visual) and motor information. Intermediate fusion achieved the highest apparent discriminative performance, reaching 97.7% accuracy on spiral drawings and 98.5% accuracy on meander drawings, whereas late fusion provided more interpretable and modular behavior, with best subject-level results of 93.94% accuracy and AUC = 0.9687 for spiral, and 92.42% accuracy with AUC = 0.9770 for meander. These findings suggest that multimodal handwriting analysis can be an effective approach for Parkinson’s disease detection on the NewHandPD dataset, although further validation on larger and independent cohorts is required.

Neurodegenerative diseases such as Parkinson’s disease (PD) and Alzheimer’s disease (AD) present a significant and growing challenge to the healthcare systems worldwide. Both conditions are progressive and often undetected early, making timely diagnosis crucial. Recently, breakthroughs in computer vision and artificial intelligence have enabled the development of non-invasive and cost-effective screening and decision-making tools, allowing for earlier detection of the disease. This review serves as a comprehensive guide, providing structured insights into computational research methods for automated detection of PD and AD, with focus on handwriting analysis as a subtle behavioral biomarker of neurological impairment. A range of methodologies is examined, including static and dynamic handwriting assessment, feature engineering procedures, deep learning and classical ML-based approaches. The analysis emphasizes the most effective methods, the handwriting features found to be most revealing, the datasets most used in the literature, and the performance levels reported for each disease. Several studies report that transfer learning based on convolutional neural networks and transformer-based architectures for Parkinson’s disease diagnosis is regularly able to achieve high accuracy, frequently above 95% on benchmark datasets. In contrast, Alzheimer’s disease research is progressively benefitting from multimodal approaches combining kinematic and spatial handwriting features to capture cognitive and motor changes. Structured summaries of publicly available handwriting datasets are provided, and critical advancements, ongoing challenges, and future research priorities are discussed. The integration of insights across the studies, through this work, aims to assist researchers and clinicians in the development and translation of handwriting-based, AI-guided diagnostic tools for neurodegenerative diseases.

This study investigates the use of offline handwriting analysis for the automated diagnosis of Parkinson’s disease (PD) using deep learning techniques. A convolutional neural network (CNN) was designed to extract discriminative spatial features from handwritten images and classify subjects as either PD patients or healthy controls. The model was evaluated on four publicly available datasets—HandPD, NewHandPD, PaHaW, and UCI—representing a diverse range of handwriting patterns and acquisition conditions. The proposed CNN achieved 100% accuracy on the smaller UCI dataset and 94.74% accuracy on the larger NewHandPD dataset. To overcome dataset imbalance and limited sample diversity, various data augmentation strategies were applied, leading to a notable increase in overall performance, with accuracies exceeding 97% on larger datasets. These results demonstrate that offline handwriting analysis, supported by deep CNN architectures and data augmentation, offers a promising, non-invasive, and cost-effective approach for early PD diagnosis and potential continuous monitoring. Furthermore, this study aligns with broader advances in AI-assisted medical diagnostics, reinforcing the role of machine learning and image-based analysis in healthcare applications.

Writer identification from handwriting is still considered to be challenging task due to homogeneous vision comparing writer of handwritten documents. This paper presents a new method based on two LBPs kinds: Complete Local Binary Patterns (CLBP) and Local Binary Pattern Variance (LBPV) for extracting the features from handwriting documents. The feature vector is then normalized using Probability Density Function (PDF). Classifications are based on the minimization of a similarity criteria based on a distance between two features vectors. A series of evaluations using different combinations of distances metrics are realized high identification rates which are compared with the methods that are participated in the ICDAR 2013 competition.

Gender Classification from handwriting is still considered to be challenging due to homogeneous vision comparing male and female handwritten documents. This paper presents a new method based on Cloud of Line Distribution (COLD) and Hinge feature for distinguishing the gender from handwriting. The SVM classifier combination decides the assigned class based on the maximum of the two decisions values resulting from COLD and Hinge feature. The proposed approach is evaluated on the standard QUWI dataset and following the framework protocol described in the ICFHR 2016 competition. Obtained results are promising regarding the classification rates announced in the literature.