A Fusion-Driven Machine learning Approach for Parkinson’s

Disease Detection & Staging in Neurodegenerative Disorder

Project (KIT-753)
Presented by Under the guidance of
Arun Kumar Singh(2107360130015) Mr. Manjeet Jaiswar
Imran Khan(2107360130033) Assistant Professor
Deepak Saroj(2107360130021) Department of Information Technology
Arpit Singh(2107360130013) Rajkiya Engineering College Azamgarh
 Table of Contents
 Introduction
 Objective of the Project
 Problem Statement
 Literature Review
 Tools & Techniques
 Dataset Used
 Methodology
 Workflow Diagram
 Algorithms
 Ensemble Learning
 UPDRS
 Result
 References
 Introduction

● Parkinson’s disease is a progressive neurodegenerative disorder that primarily affects

movement control, leading to tremors bradykinesia, rigidity and postural instability.
● Staging of Parkinson’s disease helps to assess the progression of the disorder, guiding treatment
plans and clinical decisions.
● Neurodegenerative disorders are a group of conditions characterized by the progressive loss of
structure or function of neurons (nerve cells) in the brain.
Stages of Parkinson’s Disease

Fig.1 Stages of Parkinson’s Disease

 Objective of the Project

• Develop a multimodal system integrating MRI, speech, handwriting, and wearable data.

• Detection of Parkinson’s Disease.

• Staging of Parkinson’s Disease.

• Utilize ensemble learning to combine predictions from multiple modals for improved accuracy.

• Support research on Parkinson’s progression and treatment through advanced analytics.

 Problem Statement
The project addresses the critical challenge of early and accurate detection, precise staging, and
effective management of Parkinson’s disease by leveraging multimodal data sources (MRI, speech,
handwriting and wearable Dataset) and advanced machine learning techniques. Current approaches
rely heavily on subjective clinical evaluations and underutilize the potential of integrated data, leading
to delays and a lack of scalability. This project aims to develop a unified, interpretable, and scalable
diagnostic framework that improves detection accuracy, enables granular staging, and supports real-
time symptom monitoring, bridging existing gaps in Parkinson’s disease diagnosis and treatment.
 Literature Review
Aspect Paper 1 Paper 2 Paper 3

Title Enhancing Parkinson’s Analysis of ML Prediction of Parkinson’s

Disease Detection and Algorithms to diagnose Disease using ML
Diagnosis Parkinson’s Disease Techniques

Publishing Year 2024 2023 2016

Objective Analyze advanced AI- Survey integrative Develop machine learning

based methods for early approaches across models using biomarkers
and accurate diagnosis of modalities (e.g., and non-motor symptoms
Parkinson’s Disease (PD). neuroimaging, EEG, for early prediction of
speech) for PD detection. PD.

Limitations Need for larger datasets Ethical concerns and Dependency on

and better feature challenges in integrating biomarkers and limited
selection to reduce multimodal data into generalizability of models
overfitting and clinical workflows. due to dataset size
imbalances.
 How our project differs from other Existing Projects
Aspects How our project differs

Integration of Multimodal Data Existing projects often use individual datasets (e.g., MRI,
speech, or handwriting). Our project focuses on fusion-driven
approaches that integrate multiple modalities for a holistic
diagnostic framework. This can address research gaps in
combining data sources for improved accuracy and robustness.

Advanced Fusion Techniques Our project leverage fusion-driven strategies (ensemble

learning) to combine insights from different modalities,
capturing complementary patterns.

Real-World Validation Many projects are limited to theoretical datasets. Our project
emphasize validation with real-world clinical data and robust
testing across demographic groups.
 Tools & Techniques
Category Tools and Techniques

Data Collection Google Dataset Search ,Kaggle ,

UCI-ML Repository.

Data Preprocessing Brain MRI Data, Speech Data , Handwriting Data.

Machine Learning Models K-NN Algorithm, Decision Tree , SVM .

Fusion Techniques Ensemble Learning Model

Staging Evaluation UPDRS System to determine the stage of the

Parkinson
 Dataset Used
Data Source Data Description Number of Data
Dataset

Brain MRI Structural patterns in 5536

brain images for
MRI scans Parkinson’s detection.

Hand Spiral Drawing Spiral drawings Hand-drawn spirals 20000

indicating motor
dysfunction (tremors,
irregularities).

Speech Vocal impairments in 20000

Speech recording speech indicative of
Parkinson’s disease.
 Methodology
Step 1 • Data Collection

Step 2 • Data preprocessing

Step 3 • Model Development

• Model Training,
Step 4 Optimization and Evaluation

• Model
Step 5 Deployment
Workflow
Diagram

fig.2
 Implementation
1. Data Collection
● MRI Data
● Speech Data.
● Handwriting Data

2. Data Preprocessing
● MRI Data:
○ Resizing the MRI images to a fixed size (80x80 pixels).
○ Grayscale conversion and normalization to standardize image intensity values.
● Speech Data:
○ Removing Dataset Noises.
○ Feature extraction.
● Handwriting Data:
○ Converting spiral drawings to numpy array to remove noises and ensure consistency in analysis.
○ Extracting features such as tremor patterns and drawing irregularities.
 Algorithm Models
Decision Tree SVM KNN

• Decision Tree is for analyzing Voice • SVM is for analyzing Hand Writing • Decision Tree is for analyzing MRI
Dataset. (Spiral) dataset. Dataset.
• Our Decision -based model achieved • Our SVM -based model achieved an • Our DT -based model achieved an
an accuracy of 96%. accuracy of 89%. accuracy of 97%.

Fig.3 Confusion Matrix Fig.3 Confusion Matrix Fig.3 Confusion Matrix

Fig.3 Accuracy Vs Kernal Fig.5 Model Accuracy vs K

Ensemble learning

• In our project Ensemble Learning technique is

used, where we combine multiple models to
improve the accuracy of Parkinson’s disease
prediction.

• We used the following models:

• K-Nearest Neighbors (KNN)
• Decision Tree
• Support Vector Machine (SVM)
fig.2
 Unified Parkinson’s Disease Rating
Scale (UPDRS)
• UPDRS (Unified Parkinson’s Disease Rating Scale) is a clinical tool used to evaluate the
severity and progression of Parkinson’s Disease.
• It assesses motor skills and daily activities through a scoring system.
• In our project, we used 29 parameters, each rated on a scale from 0 to 4 based on severity.
We sum up all the scores to get a total UPDRS value.
• Based on this total, we classify the stage of Parkinson’s disease.
 Final Result

Parkinson’s Not Detected Parkinson’s Detected

References
Sivaranjini, M., Sharma, A. K. (2024). Neuroimaging analysis for pathophysiological alterations in Parkinson's disease
diagnosis. International Conference on Machine Learning and Data Engineering, pp. 1–6.

Liaqat Ali, R. A. Rahman, A. A. Khan (2024). Chi2-Adaboost: A sequential learning system for Parkinson’s detection using
handwriting data. IEEE Access, Volume 9, pp. 34567–34578.

Zengyi Qin, J. Wang, S. Chen (2024). CNN-based S-Net for assessing tremor severity using sEMG signals in Parkinson’s
patients. IEEE Transactions on Neural Systems and Rehabilitation Engineering, Volume 32, No. 4, pp. 987–994.

Lerina Aversano, A. Gallo, M. Romano (2024). Deep learning for gait analysis: Classification and severity evaluation in
Parkinson’s disease. Sensors, Volume 24, No. 2, pp. 134–142.

Gayathri Nagasubramanian, R. Bhat, K. V. Prasad (2024). Speech-based Parkinson’s detection using HMM and speech
algorithms. IEEE International Conference on Speech Signal Processing, pp. 567–574.

Catherine Taleb, F. Weiss, J. Becker (2024). Online handwriting analysis for Parkinson’s detection using CNN models. IEEE
Transactions on Biomedical Engineering, Volume 71, No. 3, pp. 1004–1015.
 References (Cont.)
Manan Binth Taj Noor, H. Faisal, T. Khan (2024). Deep learning approaches for MRI-based identification of neurological
disorders. International Conference on Neural Information Processing, pp. 234–241.

Mohamed Shaban, K. Al-Otaibi, H. Al-Harbi (2024). Resting-state EEG data classification using artificial neural networks for
Parkinson’s disease diagnosis. IEEE Access, Volume 12, pp. 5484–5494.

Sura Mahmood Abdullah, L. Hussain, F. Ahmad (2024). Transfer learning for handwriting analysis using NewHandPD
dataset. IEEE Transactions on Medical Imaging, Volume 43, No. 5, pp. 1120–1130.

Chee-Keong Kwoh, Hwee-Ling Sim, Han Yu (2021). A novel feature selection method for Parkinson's disease detection using
handwritten dynamics. IEEE Transactions on Systems, Man, and Cybernetics: Systems, Volume 43, No. 2, pp. 525–535.

Tobias Blanke, Martin Fuchs, Wolfgang Maass (2021). Combining neuroimaging and machine learning for Parkinson’s
disease prediction. Frontiers in Neuroscience, Volume 15, pp. 45–60.
Thank You