NeuroSense: AI-Powered Multimodal Screening for Early-Stage Dementia

Abstract

Dementia is a progressive neurological disorder that affects memory, cognition, behaviour, and daily functioning. Early detection is essential for effective intervention and improved patient outcomes. This paper proposes NeuroSense, an AI-powered multimodal system for early-stage dementia screening by integrating speech analysis, EEG signal processing, handwriting analysis, facial emotion recognition, and cognitive assessment. The system collects data from multiple modalities and processes it through preprocessing, feature extraction, and machine learning-based classification. Speech features such as MFCC, pitch, and pause duration, EEG frequency bands, handwriting patterns, and facial expressions are analysed independently and combined using a weighted decision fusion approach. The system is implemented using a React-based frontend and a Flask backend for real-time interaction and analysis. Experimental results show that the proposed model achieves an overall accuracy of 83%, with speech analysis contributing the highest performance among all modalities. The system provides a non-invasive, cost-effective, and interpretable solution for early dementia detection, supporting healthcare professionals in clinical decision-making.

Introduction

The text describes NeuroSense, an AI-powered multimodal system developed for early dementia detection. Dementia is a progressive neurological disorder that affects memory, cognition, behavior, language, and daily functioning, creating significant challenges for patients and caregivers. Traditional diagnosis methods rely on clinical assessments, cognitive tests, and imaging techniques, which are often expensive, time-consuming, and require medical specialists.

To address these limitations, the proposed system uses Artificial Intelligence (AI) and Machine Learning (ML) to analyze multiple indicators of cognitive decline. NeuroSense integrates facial emotion recognition, speech analysis, EEG signal analysis, handwriting analysis, and cognitive assessments to improve detection accuracy. Speech and facial analysis help identify communication and emotional changes, while EEG and handwriting analysis detect neurological and motor impairments.

The literature survey reviews existing dementia prediction approaches using speech processing, MRI analysis, multimodal attention networks, and cognitive tasks. Although many studies achieved high accuracy, they often suffer from limitations such as high computational cost, dependency on specific languages, small datasets, expensive imaging requirements, and lack of scalability.

The proposed NeuroSense framework follows a structured pipeline consisting of:

• Data acquisition from speech, EEG, handwriting, facial expressions, and cognitive tests.
• Data preprocessing to remove noise and standardize inputs.
• Feature extraction using indicators like MFCC, brainwave frequencies, facial landmarks, and handwriting patterns.
• Individual modality analysis to generate dementia probability scores.
• Decision fusion that combines outputs from all modules using weighted scoring.
• Risk classification into normal, mild cognitive impairment (MCI), or high dementia risk categories.

The system is implemented using a React frontend and Flask backend, providing a user-friendly web interface and automatic PDF report generation. Experimental results show that the multimodal approach achieved 83% accuracy, with improved precision, recall, and reliability compared to single-modality systems. The project demonstrates the potential of AI-based, non-invasive, cost-effective healthcare solutions for early dementia screening and timely medical intervention.

Conclusion

The NeuroSense multimodal dementia detection system demonstrates that integrating EEG, speech, handwriting, facial expressions, and cognitive assessments significantly enhances early-stage dementia screening compared to single-modality approaches. The use of XGBoost, MobileNet, CNN, along with a rule-based automated PDF reporting system, improves classification accuracy and supports practical clinical and remote monitoring applications. For future work, the system can be further enhanced by incorporating additional modalities such as eye- tracking and gait analysis, which may capture subtle cognitive and motor changes. Advanced deep learning techniques, including transformers, can be applied to improve feature extraction and predictive performance. Developing real-time wearable or mobile monitoring systems would enable continuous assessment, while personalized risk models could provide individualized insights. Finally, conducting large-scale clinical trials will help validate scalability, robustness, and real-world applicability, ensuring the system can support early detection and long-term cognitive health monitoring effectively.

References

The NeuroSense multimodal dementia detection system demonstrates that integrating EEG, speech, handwriting, facial expressions, and cognitive assessments significantly enhances early-stage dementia screening compared to single-modality approaches. The use of XGBoost, MobileNet, CNN, along with a rule-based automated PDF reporting system, improves classification accuracy and supports practical clinical and remote monitoring applications. For future work, the system can be further enhanced by incorporating additional modalities such as eye- tracking and gait analysis, which may capture subtle cognitive and motor changes. Advanced deep learning techniques, including transformers, can be applied to improve feature extraction and predictive performance. Developing real-time wearable or mobile monitoring systems would enable continuous assessment, while personalized risk models could provide individualized insights. Finally, conducting large-scale clinical trials will help validate scalability, robustness, and real-world applicability, ensuring the system can support early detection and long-term cognitive health monitoring effectively.

Copyright

Copyright © 2026 Dr. T. Agitha, Shalini S, Sadha C, Varsha S S. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.