Dynamic Facial Expressions Analysis Based Parkinson's Disease Auxiliary Diagnosis

arXiv ID: 2512.09276v1

Published: 2025-12-10

Authors: Xiaochen Huang, Xiaochen Bi, Cuihua Lv, Xin Wang, Haoyan Zhang, Wenjing Jiang, Xin Ma, Yibin Li

Categories: cs.CV

Relevance Score: 0.98 / 1.00

View on arXiv Download PDF

Summary

This paper introduces a novel dynamic facial expression analysis method for auxiliary Parkinson's Disease (PD) diagnosis, targeting hypomimia through the analysis of reduced facial expressivity and rigidity. Utilizing a multimodal network leveraging CLIP for visual-textual integration and temporal dynamics, followed by an LSTM-based classifier, the method achieves 93.1% accuracy. This approach offers a more convenient and efficient diagnostic tool, outperforming existing in-vitro PD diagnostic techniques.

Medical Relevance

This research is highly relevant as it proposes a non-invasive, accessible, and efficient method for the auxiliary diagnosis of Parkinson's Disease. By leveraging observable facial symptoms like hypomimia, it can potentially facilitate earlier detection and improve the diagnostic experience for patients, which is crucial for managing this progressive neurodegenerative disorder.

AI Health Application

The AI application involves a multimodal facial expression analysis network, leveraging the CLIP architecture to integrate visual and textual features, and preserving temporal dynamics. This network extracts expression intensity features which are then processed by an LSTM-based classification network to provide an auxiliary diagnosis for Parkinson's Disease based on analyzing hypomimia (reduced facial expressivity and rigidity).

Key Points

A dynamic facial expression analysis method is proposed for auxiliary Parkinson's Disease (PD) diagnosis.
The method specifically targets hypomimia, a characteristic PD symptom, by analyzing reduced facial expressivity and facial rigidity.
A multimodal facial expression analysis network, based on the CLIP architecture, is developed to extract expression intensity features.
This network integrates visual and textual features while preserving the crucial temporal dynamics of facial expressions.
Extracted expression intensity features are then processed and fed into an LSTM-based classification network for final PD diagnosis.
The proposed technique achieved a high diagnostic accuracy of 93.1%, demonstrating superior performance compared to other in-vitro PD diagnostic approaches.
The method offers a more convenient and accessible detection strategy for potential PD patients, aiming to improve their diagnostic experience.

Methodology

The method involves a multimodal facial expression analysis network built upon the CLIP architecture to integrate visual and textual features, ensuring the preservation of temporal dynamics. This network extracts expression intensity features from patients performing various facial expressions. Subsequently, these features are fed into an LSTM (Long Short-Term Memory)-based classification network for final Parkinson's Disease diagnosis.

Key Findings

The primary finding is that the proposed dynamic facial expression analysis method achieved an accuracy of 93.1% in diagnosing Parkinson's Disease. This performance significantly outperforms other existing in-vitro PD diagnostic approaches, demonstrating its efficacy and potential as a superior diagnostic tool.

Clinical Impact

This technique has the potential for significant clinical impact by offering a more convenient, non-invasive, and accessible method for the auxiliary diagnosis of Parkinson's Disease. It could enable earlier screening and detection of potential PD patients, streamline the diagnostic pathway, and ultimately improve the patient diagnostic experience by reducing reliance on more invasive or time-consuming traditional methods.

Limitations

Not explicitly mentioned in the abstract.

Future Directions

Not explicitly mentioned in the abstract.

Medical Domains

Neurology Movement Disorders Diagnostic Medicine Geriatrics Digital Health

Keywords

Parkinson's Disease Hypomimia Facial Expression Analysis CLIP Architecture LSTM Neurodegenerative Disorder Auxiliary Diagnosis Computer Vision

Abstract

Parkinson's disease (PD), a prevalent neurodegenerative disorder, significantly affects patients' daily functioning and social interactions. To facilitate a more efficient and accessible diagnostic approach for PD, we propose a dynamic facial expression analysis-based PD auxiliary diagnosis method. This method targets hypomimia, a characteristic clinical symptom of PD, by analyzing two manifestations: reduced facial expressivity and facial rigidity, thereby facilitating the diagnosis process. We develop a multimodal facial expression analysis network to extract expression intensity features during patients' performance of various facial expressions. This network leverages the CLIP architecture to integrate visual and textual features while preserving the temporal dynamics of facial expressions. Subsequently, the expression intensity features are processed and input into an LSTM-based classification network for PD diagnosis. Our method achieves an accuracy of 93.1%, outperforming other in-vitro PD diagnostic approaches. This technique offers a more convenient detection method for potential PD patients, improving their diagnostic experience.