Parkinson’s disease presents significant diagnostic challenges and is often delayed due to its complex nature and reliance on subjective assessment. Florida International University and University of South Florida researchers Md Mezbahul Islam, John Michael Templeton, Masrur Sobhan, along with Christian Poellabauer, Ananda Mohan Mondal and others, are using SCOPE-PD, a new explainable artificial intelligence framework, to address this critical need. This study uniquely integrates both subjective patient reports and objective clinical measurements to not only predict Parkinson’s disease with a high accuracy of 98.66% using a random forest algorithm, but also to provide transparent and interpretable insights into the factors driving the prediction, such as tremor, bradykinesia, and facial expressions, ultimately enabling more personalized and informed medical decision-making.
Multimodal machine learning predicts Parkinson’s disease progression with high accuracy and interpretability, offering potential for personalized treatment planning
Scientists developed SCOPE-PD, an explainable artificial intelligence framework designed to improve prediction of Parkinson’s disease by integrating both subjective patient reports and objective clinical assessments. This research addresses the critical need for early and accurate diagnosis of Parkinson’s disease, a neurodegenerative disease that currently affects 11.8 million people worldwide and is predicted to rise to more than 25 million by 2050.
Using research data from the Parkinson’s Disease Progression Markers Initiative, the research team made significant progress by building a multimodal predictive framework that combines self-reported outcomes and expert-rated testing. Several machine learning techniques were applied to this data, and in the end, the Random Forest algorithm showed the highest accuracy of 98.66% when leveraging a combination of features.
The interpretability of the model was then examined using SHAP-based analysis. SHAP-based analysis is a technique that clarifies how predictions are made and addresses key barriers to clinical implementation of machine learning models. This study establishes a clear relationship between specific clinical characteristics and predicted risk, providing a more transparent and reliable diagnostic tool for clinicians and patients.
This study revealed that tremor, bradykinesia, and facial expressions were the top three contributing features of the MDS-UPDRS test in predicting Parkinson’s disease. By quantifying the contribution of both patient-specific and cohort-level characteristics, SCOPE-PD enables intuitive statements about how individual factors influence the probability of disease, such as “this characteristic increased the probability of PD by +0.05.”
This innovation not only improves diagnostic accuracy, but also fosters trust and understanding, paving the way for future explainable AI frameworks in neurodegenerative disease research and precision medicine. This study demonstrates the value of integrating diverse data types to improve diagnosis and identifies key features that influence prediction. This research opens the possibility of personalized health decisions and a more nuanced understanding of the progression of Parkinson’s disease, which could lead to earlier intervention and improved patient outcomes.
Development and validation of a multimodal Parkinson’s disease prediction framework using explainable artificial intelligence is essential for early diagnosis
Scientists developed SCOPE-PD, an explainable AI-based predictive framework for Parkinson’s disease, integrating both subjective and objective clinical assessments to facilitate personalized medical decision-making. The research team collected data from the Parkinson’s Disease Progression Markers Initiative (PPMI) study and built a multimodal predictive framework that leverages a comprehensive dataset of clinical information.
The best model was then selected based on interpretability and predictive performance, and several machine learning techniques were applied to this data. To rigorously assess model interpretability, this study pioneered the use of SHAP-based analysis, a method for explaining the output of machine learning models.
This approach allowed researchers to understand the contribution of each feature to the final prediction, providing insight into the underlying factors driving model decisions. The experiment employed a random forest algorithm, which achieved the highest accuracy of 98.66 percent when utilizing a combination of features from both subjective and objective test data.
The system significantly improves diagnostic accuracy by leveraging a complete picture of patient information, surpassing traditional methods that rely solely on subjective or objective assessments. Detailed analysis revealed that tremor, bradykinesia, and facial expressions were identified as the top three contributing features of the MDS-UPDRS test in predicting Parkinson’s disease.
This innovative approach enables the identification of key clinical indicators, which may facilitate earlier and more accurate diagnosis and ultimately improve patient outcomes. This study represents an important step toward transforming machine learning into a practical clinical tool for neurodegenerative disease management.
Random forest prediction for Parkinson’s disease with integrated clinical and patient-reported data shows promising results
Scientists achieved 98.66% accuracy in predicting Parkinson’s disease using a new machine learning framework called SCOPE-PD. This study integrated both subjective patient reports and objective clinical assessments collected from the Parkinson’s Disease Progression Markers Initiative (PPMI) study. Experiments revealed that the random forest algorithm performs best when leveraging features from both data types in combination, resulting in significantly improved predictive power.
According to the data, tremor, bradykinesia, and facial expressions were identified as the top three features contributing to the MDS-UPDRS test in predicting Parkinson’s disease. The team used SHAP-based analysis to measure the importance of these features and provide insight into the model’s decision-making process.
The results demonstrate a clear association between these specific clinical indicators and accurate identification of the disease. Tests demonstrate the framework’s ability to harmonize traditional clinical practice with new computational techniques to improve diagnosis. In this study, we combined subjective and objective features to train and compare state-of-the-art classification algorithms.
Measurements confirm that integrating different measurements from the PPMI repository improves predictive accuracy and achieves significant improvements over traditional diagnostic methods. This breakthrough provides clinically relevant explanations for each prediction, leveraging advanced SHAP tools to build trust and understanding.
Scientists note that identifying key features such as tremor and bradykinesia provides valuable insight for clinicians. This approach not only improves diagnostic accuracy but also facilitates personalized health decision-making and establishes a new standard for explainable AI in neurodegenerative disease research. This study highlights the potential for earlier and more accurate diagnosis of Parkinson’s disease, potentially reducing the current annual cost in the United States of more than $50 billion.
Integrating subjective and objective data improves prediction accuracy for Parkinson’s disease
Scientists developed SCOPE-PD, an explainable machine learning framework designed to predict Parkinson’s disease by integrating both subjective and objective clinical assessments. This study demonstrated that a random forest algorithm utilizing a combination of data types achieved 98.66 percent accuracy in identifying Parkinson’s disease.
The main features contributing to this prediction, identified through SHAP-based analysis, were tremor, bradykinesia, and facial expressions measured by the MDS-UPDRS test. The novelty of this study lies in the integration of subjective and objective clinical data within a single interpretable predictive model, rather than exceeding existing accuracy benchmarks.
This framework aims to bridge the gap between traditional diagnostic methods and modern machine learning techniques and may provide a valuable screening tool for early detection and personalized treatment planning. However, the authors acknowledge that there are limitations, such as the absence of external validation datasets, which limit the assessment of the model’s robustness and real-world applicability.
Future research will focus on incorporating multisite data from additional Parkinson’s disease databases for external validation and clinical trials to evaluate the model’s impact on diagnostic accuracy and patient care. Although SCOPE-PD is not immediately clinically deployable, it represents a methodological advance, provides an explainable approach to risk estimation, and has the potential to identify at-risk individuals before significant symptoms develop. Interpretability of the model is facilitated by a random forest algorithm and is suitable for assisting clinicians in making evidence-based diagnoses.
👉 More information
🗞 SCOPE-PD: Explainable AI for subjective and clinical objective measurements of Parkinson’s disease for accurate decision making
🧠ArXiv: https://arxiv.org/abs/2601.22516
