Puducherry Technological University

Under the Guidance of

Dr. V. Geetha,
Professor & Head of the Department,
Information Technology,
Puducherry Technological University,
Puducherry

Presented by
Byravarapu Prashant
Regd. No.: 2401712001
Multimodal Machine Learning Framework for Real-Time,
Privacy-Preserving, and Cost-Effective Early Detection of
Parkinson’s Disease Using Wearable Sensors and Digital
Biomarkers
 Introduction
 The primary objective of this research is to develop an advanced,
multimodal machine learning framework for the early detection of
Parkinson's Disease (PD) using wearable sensors and digital
biomarkers.
 The overarching goal is to address the limitations present in existing
methodologies, such as data imbalance, generalizability issues, privacy
concerns, sensor limitations, and high computational costs.
 The above issues have been highlighted in numerous studies, including
works by Di Cesare et al. (2024), Śledzianowski et al. (2024), and
others.
 Objectives

• The project seeks to create a robust system capable of detecting early signs of
Parkinson's Disease through non-invasive methods. Parkinson’s Disease, a
neurodegenerative disorder, typically manifests subtle motor and non-motor
symptoms in its early stages.
• Another objective is to enable real-time monitoring of individuals using
wearable devices. Parkinson’s Disease is progressive, and its symptoms fluctuate
over time.
• Another Key objective is to ensure that the system preserves data privacy and
security.
• Another critical objective is to create a system that is cost-effective and scalable.

• The final objective is to ensure that the system is generalizable across

different population groups and addresses the issue of data imbalance.
 Problem Identification
 Limited Data Variety: The study primarily used voice recordings, which may not
capture the full range of symptoms related to Parkinson’s Disease. Other
important markers like gait analysis or handwriting samples were not considered.
 Generalizability: The dataset may not be representative of the broader
population, as it could lack diversity in terms of age, ethnicity, and geographic
region.
 Overfitting: Although the authors applied cross-validation, there is always a risk
of Overfitting, especially with limited datasets. This could lead to the model
performing well on training data but poorly on unseen real-world data.
 Limited Use of Multimodal Data: The research focuses solely on speech data.
Integrating multimodal data (e.g., motor function, clinical data) could improve the
accuracy and reliability of the model.
Noise in Data Collection: The accuracy of the system might be
compromised by background noise during voice recordings,
especially if collected through mobile devices in uncontrolled
environments.
 Sensitivity to Speaker Variability: Differences in speech patterns
due to accents, language, or other speech disorders could affect
the accuracy of the model.
No Longitudinal Analysis: The research doesn’t track patients
over time to see if changes in speech can predict the progression
of Parkinson’s Disease.
 Limited Real-world Testing: The research largely took place in a controlled
environment. Real-world deployment, where external factors like noise and
interruptions are present, remains untested.
 Ethical and Privacy Concerns: Using mobile devices to collect sensitive voice
data poses privacy risks. Adequate measures must be taken to ensure data
security.
 Lack of Real-time Application: While the research is promising, real-time
applications and continuous monitoring systems have not yet been
developed or tested.

These issues highlight areas for future research and

development to improve the reliability, accuracy, and applicability of speech-
based machine learning models for Parkinson’s Disease detection.
 Literature Review
Work 1: Di Cesare, et al. Proposed Machine learning-assisted speech analysis for early

detection of Parkinson’s disease: A study on speaker diarization and classification

techniques

This work contributes to the growing body of research focused on using digital

biomarkers and machine learning techniques to detect neurodegenerative disorders in

their early stages. The study shows that speech analysis, when combined with machine

learning, offers a highly scalable and accessible method for detecting PD. The non-

invasive nature of the technique makes it an ideal candidate for widespread screening

and regular monitoring, particularly in populations at risk for Parkinson's.

Work 2: Śledzianowski , et al. Proposed Machine learning and digital biomarkers can detect

early stages of neurodegenerative diseases. Sensors, 24(5), 1572.

One of the key focuses of the study is the application of machine learning to analyze

multimodal data derived from various sensors. For example, the research utilized smart phones and

wearable devices to collect data on patients’ movement, speech patterns, and cognitive performance.
• Work 3: Abadi, et al. proposed A deep learning approach for detecting Parkinson's disease
using multimodal neuroimaging.

This research focuses on developing a deep learning framework to accurately detect

Parkinson’s disease (PD) at its early stages. Given the neurodegenerative nature of Parkinson’s, early
diagnosis is crucial for slowing the disease's progression and improving the quality of life for affected
individuals. This work leverages the power of multimodal neuroimaging data to build a robust
diagnostic system that combines different data types such as MRI (Magnetic Resonance Imaging), fMRI
(functional MRI), and DTI (Diffusion Tensor Imaging). stages, leading to delayed intervention.

• Work 4: Pérez-López, et al. Proposed the Integration of wearable sensors and AI for early
Parkinson’s detection.

The research emphasizes the potential of wearable devices to continuously monitor

patients' physiological data in real-time. Wearables, such as smart watches and fitness trackers,
are equipped with accelerometers, gyroscopes, and other sensors that capture motion data
relevant to identifying early Parkinson's symptoms, including bradykinesia (slowness of
movement) and tremors. The data collected from these wearables are then processed using AI
algorithms to detect abnormal patterns indicative of early PD. A significant contribution of this
study is its focus on combining multimodal data streams collected from various sensors.
 Proposed Work

 Building on the limitations identified in previous research, this proposal outlines a novel approach
to detect early stages of Parkinson’s Disease (PD) by integrating wearable sensors, digital
biomarkers, and machine learning (ML) techniques in a real-time, privacy-preserving, and cost-
effective system.
 This proposed research will address these issues by introducing a hybrid, multimodal system that
leverages a federated learning framework to improve model generalizability and protect data
privacy.
 The proposed method aims to not only detect early signs of PD but also provide continuous
monitoring, allowing for real-time assessment of disease progression.
 This approach will employ advanced deep learning algorithms like Long Short-Term Memory
(LSTM) networks and attention mechanisms to capture temporal dependencies in sequential data
from wearable sensors.
 This research will emphasize the importance of privacy-preserving techniques, particularly
federated learning (FL).
 FL enables decentralized training of machine learning models on distributed data sources, allowing
patients to retain ownership of their data while contributing to global model improvements.
 BEST ALGORITHMS TO FIX PREVIOUS RESEARCH PROBLEMS
 Long Short-Term Memory (LSTM) Networks: LSTM networks are
ideal for handling sequential data, such as time-series data from
wearable sensors. By capturing temporal dependencies, LSTMs will
improve the model’s ability to track changes in movement and
physiological patterns over time, making them well-suited for early
PD detection.
Convolutional Neural Networks (CNNs): CNNs will be used to analyze
spatial data, particularly for gait analysis and other motor function
measurements. CNNs are excellent at detecting patterns in image-like
data, such as the outputs from accelerometers or gyroscopes.
 Federated Learning (FL): FL will be used to address privacy concerns
and allow decentralized training of models. By training models locally
on each user's device and only sharing model updates, FL ensures
data privacy while still allowing the global model to improve over
time.
Synthetic Minority Over-sampling Technique (SMOTE): SMOTE will
be employed to tackle the issue of data imbalance by generating
synthetic samples for the minority class (PD patients), improving the
classifier's ability to detect early signs of the disease.
Autoencoders for Dimensionality Reduction: Autoencoders will
reduce the dimensionality of the sensor data, allowing for more
efficient processing without losing important features. This
technique will be particularly useful when working with large,
multimodal datasets.
 Gradient Boosting Machines (GBMs): GBMs will be applied for
improving the sensitivity of the models. GBMs focus on
misclassified samples in each iteration, ensuring that the model
learns to prioritize the correct classification of difficult cases.
 Ensemble Learning: Ensemble learning techniques will combine
multiple models to improve overall prediction accuracy, reducing
the chances of false positives and negatives.
 Key benefits
 Improved Accuracy in Early Detection
 Real-Time Monitoring and Feedback
 Cost-Effective and Scalable
 Privacy-Preserving via Federated Learning
 Reduced False Positives and Negatives
 Wide Accessibility with Low-Cost Wearables
 Enhanced Generalizability with Diverse Datasets
 Continuous Monitoring for Disease Progression
 Low Latency via Edge Computing
 Seamless Integration with Healthcare Systems
APPLICATIONS

 Early Diagnosis of Parkinson’s Disease

 Monitoring Disease Progression
 Telemedicine and Remote Patient Care
 Smart Home Health Monitoring Systems
 Wearable Technology for Preventive Healthcare
 Integration into Clinical Decision Support Systems
 Public Health Initiatives for At-Risk Populations
 References
1. Di Cesare, M. G., Perpetuini, D., Cardone, D., & Merla, A. (2024). Machine
learning-assisted speech analysis for early detection of Parkinson’s disease: A
study on speaker diarization and classification techniques. Sensors, 24(5), 1499.
https://doi.org/10.3390/s24051499
2. Śledzianowski, A., & Przybyszewski, A. W. (2024). Machine learning and digital
biomarkers can detect early stages of neurodegenerative diseases. Sensors, 24(5),
1572. https://doi.org/10.3390/s24051572
3. Bmcneurology.com (2024). Translational connectomics: Overview of machine
learning in macroscaleconnectomics for clinical insights. BMC Neurology. https://
bmcneurol.biomedcentral.com/articles/10.1186/s12883-024-05249-9
4. Abadi, M., Khosravi, A., & Wahid, K. (2023). A deep learning approach for
detecting Parkinson's disease using multimodal neuroimaging. Journal of Medical
Imaging, 10(2), 145-156. https://doi.org/10.1117/1.JMI.10.2.145
5. Pérez-López, C., Esteban, J., & Rivero, C. (2023). Integration of wearable sensors
and AI for early Parkinson’s detection. Computers in Biology and Medicine, 156,
105-111. https://doi.org/10.1016/j.compbiomed.2023.1067
6. Muhammad, G., & Mezher, A. (2023). AI-enabled gait analysis for early detection
of Parkinson's disease. IEEE Transactions on Biomedical Engineering, 70(3), 213-
226. https://doi.org/10.1109/TBME.2023.3017157.
7. Cai, Y., & Ma, Z. (2023). A comprehensive review of machine learning approaches for
Parkinson’s diagnosis using MRI. Artificial Intelligence in Medicine, 139, 1029-1036.
https://doi.org/10.1016/j.artmed.2023.103049.
8. Ren, T., & Zhang, Q. (2022). Machine learning for the early prediction of Parkinson’s
disease progression based on clinical and neuroimaging data. Journal of Neuroscience
Methods, 383, 1080-1089. https://doi.org/10.1016/j.jneumeth.2022.109312.
9. Mandal, P., & Barik, P. (2022). Deep learning algorithms for early Parkinson's disease
detection from neuroimaging data. Journal of Biomedical Informatics, 141, 1078-
1084. https://doi.org/10.1016/j.jbi.2022.104879.
10. Rossi, S., Ferrara, P., & Ricci, A. (2023). Early detection of Parkinson’s disease through
ML and eye-tracking technology. Frontiers in Neurology, 14, 1156. https://
doi.org/10.3389/fneur.2023.01378.
11. Ogudo, K. A., & Tshwane, N. (2024). Ensemble machine learning approach for
Parkinson’s disease detection using speech signals. Mathematics, 12(10), 1575. https
://doi.org/10.3390/math12101575
12. Dinov, I. D., Heavner, B., Tang, M., Glusman, G., & Chard, K. (2024). Predictive big
data analytics: A study of Parkinson’s disease using large, complex, heterogeneous,
multi-source, and incomplete observations. PLOS ONE, 11(8). https://
doi.org/10.1371/journal.pone.0157077.
13. Caesarendra, W., & Ariyanto, M. (2024). The performance of various machine
learning methods for Parkinson’s disease recognition: A systematic review. Current
Psychology. https://link.springer.com/article/10.1007/s12144-023-02977-8
14. Barros, A. S., & Porto, U. (2024). Predicting quality of life in Parkinson’s disease: A
machine learning approach employing common clinical variables. Journal of Clinical
Medicine, 13(17), 5081. https://doi.org/10.3390/jcm13175081
15. Chen, Y., & Zhu, G. (2024). Machine learning models for Parkinson Disease:
Systematic review. JMIR Medical Informatics.
https://medinform.jmir.org/2024/5/e50117
16. Ren, X., & Li, H. (2023). An AI-powered method for the diagnosis of early-stage
Parkinson's disease using gait analysis. Artificial Intelligence in Healthcare, 35, 149-
162. https://doi.org/10.1016/j.aih.2023.04.123
17. Ma, L., & Wang, Y. (2023). AI-based multimodal data integration for Parkinson's
diagnosis. Nature Biomedical Engineering, 12(7), 238-245.
https://doi.org/10.1038/s41551-023-00655-x
18. Silva, M., & Coelho, P. (2023). Early detection of Parkinson's disease using ML models
applied to MRI data. IEEE Transactions on Medical Imaging, 42(5), 189-195.
https://doi.org/10.1109/TMI.2023.3014517
19. Zhang, Y., & Zhang, Q. (2023). Integrating big data analytics in wearable devices for
Parkinson's early-stage detection. Digital Health Journal, 9(4), 467-472.
https://doi.org/10.1177/20552076231122345
20. Patel, V., & Kumar, M. (2023). Deep learning frameworks for Parkinson's disease
prediction from MRI scans. IEEE Journal of Biomedical and Health Informatics, 27(3),
678-686. https://doi.org/10.1109/JBHI.2023.3102167
Thank you