New research has demonstrated that explainable artificial intelligence (XAI) frameworks can be significantly improved. parkinson’s disease (PD) provides clinically meaningful insight into how decisions are made while providing predictions.
PD is a progressive neurological disorder characterized by motor symptoms such as tremor and rigidity, and non-motor symptoms such as cognitive impairment and sleep disturbances.
Early diagnosis remains difficult, especially in the early stages when symptoms are subtle or overlap with other symptoms. While machine learning shows promise in aiding diagnosis, limited interpretability hinders its clinical adoption.
Explainable AI improves Parkinson’s disease predictions
In this study, researchers developed a multimodal framework that combines machine learning and XAI techniques to improve PD prediction. This model integrated disparate data sources including neuroimaging, clinical features, and both motor and non-motor symptoms, allowing for a more comprehensive assessment of disease risk.
Several machine learning algorithms were evaluated, including support vector machines, random forests, k-nearest neighbor algorithms, and decision trees. These models, combined with XAI tools such as SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explains (LIME), and ELI5, provide both global and individual-level explanations of predictions.
The results showed that the enhanced framework performed better than traditional machine learning approaches. In particular, the AdaBoost model achieved the best performance with accuracy of 93%, precision of 90%, recall of 90%, F1 score of 90%, and area under the curve of 0.95.
This represents a measurable improvement over the baseline model and highlights the added value of integrating explainability into predictive systems.
Bridging accuracy and clinical interpretability
The main strength of this framework was its ability to identify the most influential features contributing to PD prediction. By providing transparent and interpretable output, clinicians can better understand which neuroimaging markers and clinical symptoms are driving individual predictions.
This overcomes a major limitation of many artificial intelligence models, often referred to as “black boxes,” where high accuracy is achieved without a clear decision-making process. Including both local and global descriptions increases clinician confidence and offers the potential for improved integration into real-world practice.
Impact on early diagnosis and individualized care
The findings of this study suggest that XAI may play a pivotal role in promoting early PD prediction and supporting personalized treatment strategies. This approach has the potential to enable earlier intervention and better clinical decision-making by combining predictive performance and interpretability.
Further validation in larger and more diverse populations will be essential before widespread clinical implementation. Nevertheless, this study represents an important step towards accurate and clinical implementation of artificial intelligence in neurology.
reference
Mehta V et al. A multimodal explainable artificial intelligence framework for interpretable Parkinson’s disease prediction. Sci Rep. 2026;DOI:10.1038/s41598-026-47769-z.
Featured image: New Africa from Adobe Stock
