AI-POWERED ANALYSIS
Enterprise AI Analysis: Integration of fNIRS and Machine Learning for Identifying Parkinson's Disease
This study demonstrates the effectiveness of combining functional near-infrared spectroscopy (fNIRS) with machine learning to diagnose Parkinson's Disease (PD). Utilizing dual-task activities, the SVM model achieved 85% accuracy in distinguishing PD patients from controls. Key prefrontal cortex (PFC) subregions were identified as potential biomarkers, highlighting the promise of non-invasive diagnostic tools for early intervention.
Executive Impact
Leverage cutting-edge AI for precise insights. Our analysis highlights the transformative potential of advanced machine learning in medical diagnostics.
0
SVM Accuracy
0
PD Cases Correctly Classified
0
CG Cases Correctly Classified
0
AUC Score (SVM)
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Methodology
Key Findings
Conclusion & Future Work
Data Collection & Preprocessing
Data was collected using fNIRS from 28 individuals with PD and 32 age-matched controls during Timed Up and Go (TUG) tests under three conditions: simple TUG, cognitive dual-task TUG, and motor dual-task TUG. Signals underwent rigorous preprocessing including noisy channel identification, baseline correction, artefact removal, and filtering to ensure high quality. Channels with a Coefficient of Variation (CV) exceeding 7.5% were excluded. Ultimately, data from 27 participants per group were used for analysis. The PFC area, covering Dorsolateral, Ventrolateral, Medial, and Frontopolar cortices, was monitored with 24 channels (10 sources, 8 detectors).
Statistical Analysis & Feature Extraction
Repeated measures ANOVA examined effects of walking conditions on oxyhaemoglobin (ΔΗΒΟ) and deoxyhaemoglobin (ΔΗHB) concentrations in the PFC, with group and gender as between-subjects factors. Mean ΔΗΒΟ and ΔΗHB values from each channel during each task served as inputs for machine learning models. Two filter-based feature selection methods—Joint Mutual Information (JMI) and ANOVA F-statistic—were employed to identify crucial features and reduce dimensionality.
Machine Learning Models & Optimisation
Four machine learning models were implemented: Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and Extreme Gradient Boosting (XGB). Hyperparameters for each model were optimized using a grid search approach combined with ten-fold cross-validation. Model performance was evaluated using leave-one-out cross-validation (LOOCV), with metrics including accuracy, sensitivity, precision, and F1-score.
Overall Classification Performance
The SVM model, particularly with ANOVA-based feature selection and combined ΔΗΒΟ and ΔΗHB measures, achieved the highest accuracy of 0.85 ± 0.35. Dual-task activities were found to be more effective in distinguishing PD from controls compared to simple tasks. The model demonstrated strong performance in identifying individuals with PD (92% accuracy for PD cases), although slightly lower for the control group (77%). The ROC curve yielded an AUC of 0.81.
Cortical Activation Patterns
Significant differences in ΔΗHB levels were observed between PD and control groups during dual-task conditions (specifically MDTUG), with channel-wise analysis revealing key distinctions. For the PD group, highest activation was during simple TUG, followed by MDTUG, with CDTUG showing the least. Control group activation was highest during TUG and CDTUG, with MDTUG showing the least. These patterns suggest varying cognitive load and compensatory mechanisms.
Identified Biomarkers & Subregions
ANOVA feature selection identified top 11 features (specific channels and modalities) as crucial for classification. Four key channels (6, 7, 18, 19) in the Frontopolar Cortex (FPC) subregion were particularly significant, indicating its role in higher-order cognitive functions. DLPFC and VLPFC regions also contributed. Notably, no channels from the MPFC area were found within the top features. Dual-task activities, especially MDTUG and CDTUG, elicited the most discriminative activation patterns.
Implications for PD Diagnosis
The study highlights the potential of fNIRS combined with machine learning as a non-invasive, objective tool for early PD diagnosis and monitoring. The identification of specific PFC subregions and dual-task activation patterns provides valuable insights for developing targeted rehabilitation strategies and understanding disease progression. This approach complements traditional clinical assessments.
Limitations & Future Directions
Limitations include a relatively small sample size (though power-analyzed), an unbalanced sex distribution (though analysis showed no significant effect), and a classification bias towards PD cases. Future work should involve larger, more diverse cohorts, balanced sex distribution, and explore additional brain regions (e.g., PMC). Incorporating temporal dynamics, advanced deep learning models, and multimodal assessments (e.g., body motion) can further improve accuracy and generalizability.
0
SVM Accuracy with ANOVA Feature Selection
The Support Vector Machine (SVM) model, when integrated with ANOVA-based feature selection, achieved an impressive accuracy of 0.85 ± 0.35 in distinguishing Parkinson's disease (PD) patients from age-matched controls.
End-to-End fNIRS-ML Workflow for PD Detection
fNIRS Data Collection (TUG, CDTUG, MDTUG)
→
Signal Preprocessing (Noise, Artefact, Filtering)
→
Data Segmentation & Feature Extraction (ΔΗΒΟ, ΔΗΗΒ mean values)
→
Machine Learning for PD Classification (SVM, KNN, RF, XGB)
→
Hyperparameter Optimization & Feature Selection (ANOVA F-statistic, JMI)
→
Biomarker Identification & PD-CG Classification
| Feature | Dual-Task Conditions (CDTUG, MDTUG) | Simple Task (TUG) |
|---|---|---|
| PFC Activation Patterns |
|
|
| Cognitive Load & Executive Function |
|
|
| Biomarker Identification |
|
|
| Classification Accuracy |
|
|
PFC Subregions as Key Biomarkers for PD Detection
Specific subregions within the Prefrontal Cortex (PFC) exhibit distinct activation patterns that serve as critical biomarkers for distinguishing Parkinson's disease (PD) from age-matched controls. Our analysis highlights the importance of the Frontopolar Cortex (FPC), Dorsolateral Prefrontal Cortex (DLPFC), and Ventrolateral Prefrontal Cortex (VLPFC).
Frontopolar Cortex (FPC)
Four key channels (6, 7, 18, and 19) in the FPC subregion were identified as highly discriminative. The FPC is crucial for higher-order cognitive functions like attention allocation and decision-making, which are impaired in PD, especially during dual-task activities. Channel 7 in FPC showed the highest ΔΗΒΟ level for controls and lowest for PD during CDTUG.
Dorsolateral Prefrontal Cortex (DLPFC)
Channels in the DLPFC, such as Channel 10, displayed higher ΔΗΒΟ concentration in the control group compared to PD during CDTUG. This subregion is known to be impacted in PD and its differential activity further supports its role as a biomarker.
Ventrolateral Prefrontal Cortex (VLPFC)
Channels in the VLPFC (e.g., Channel 22, 23) showed higher activation in controls during simple TUG and lower ΔΗHB activation in PD during MDTUG, indicating varying compensatory mechanisms or cognitive effort between groups.
Absence in Medial Prefrontal Cortex (MPFC)
Notably, no channels from the MPFC area were found within the top 11 features, suggesting its potentially lesser role in distinguishing PD in the contexts studied here compared to other PFC subregions.
These findings underscore the value of specific PFC subregions as localized and highly informative biomarkers, enhancing our understanding of PD-related neural activity during complex motor-cognitive tasks.
Calculate Your Potential AI ROI
Estimate the significant return on investment your enterprise could achieve by integrating our AI solutions, tailored to your operational specifics.
Estimated Annual Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A clear, phased approach to integrating AI into your enterprise, ensuring maximum impact with minimal disruption.
Phase 1: Discovery & Strategy
In-depth analysis of current workflows, identification of AI opportunities, and development of a tailored strategy aligned with your business objectives. This phase involves stakeholder interviews, data assessment, and ROI projections.
Phase 2: Solution Design & Prototyping
Designing the AI architecture, selecting appropriate models, and developing initial prototypes. We focus on creating a scalable and robust solution that integrates seamlessly with your existing infrastructure.
Phase 3: Development & Integration
Full-scale development of the AI solution, rigorous testing, and phased integration into your operational environment. Our team ensures smooth deployment and minimal impact on daily operations.
Phase 4: Training & Optimisation
Comprehensive training for your team, ongoing performance monitoring, and continuous optimization of the AI models to ensure sustained efficiency and evolving business needs are met.
Ready to Transform Your Enterprise with AI?
Don't let complexity hold you back. Our experts are ready to guide you through every step of your AI journey. Book a personalized consultation today.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat