TY  - JOUR
AU  - Raza, Aamir
AU  - Ali, Aamir
AU  - Kumar, Aashesh
AU  - Fatima, Nikhat
AU  - Ali, Misbah
PY  - 2026
DA  - 2026/03/12
TI  - Bridging Predictive Modeling and Clinical Interpretability: An Explainable AI Approach to Parkinson’s Disease Detection
JO  - Biomedical Informatics and Smart Healthcare
T2  - Biomedical Informatics and Smart Healthcare
JF  - Biomedical Informatics and Smart Healthcare
VL  - 2
IS  - 1
SP  - 20
EP  - 37
DO  - 10.62762/BISH.2026.470997
UR  - https://www.icck.org/article/abs/BISH.2026.470997
KW  - Parkinson’s disease
KW  - explainable artificial intelligence
KW  - machine learning
KW  - deep learning
KW  - feature selection
KW  - clinical decision support
AB  - Parkinson’s disease (PD) is the second most common neurodegenerative disorder worldwide, predominantly affecting older adults. Early detection is crucial, as subtle motor and non-motor symptoms frequently overlap with other conditions, often resulting in delayed diagnosis. Many existing models rely on costly and less accessible imaging modalities such as MRI or PET scans, limiting their applicability in resource-constrained settings where only routine clinical data are available. This study develops interpretable AI models for early PD detection using structured clinical variables, incorporating feature selection techniques. Feature selection was conducted via Random Forest (RF) importance ranking combined with SelectKBest statistical scoring, retaining the most informative predictors for modeling. Five classifiers were implemented in parallel: Support Vector Machine (SVM), K-Nearest Neighbors (KNN), RF, Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) network. Model performance was assessed using accuracy, precision, recall, F1-score, and ROC-AUC metrics. The RF model achieved the highest accuracy of 92.72% (ROC-AUC: 0.968), with CNN and SVM showing competitive performance. LSTM exhibited balanced sensitivity and specificity, while KNN demonstrated relatively lower recall. To improve clinical interpretability, LIME was applied to each model to produce instance-level explanations, consistently highlighting tremor severity, motor impairment, cognitive scores, and age as key influential features. These results demonstrate that structured clinical variables alone can enable reliable PD detection without dependence on imaging. Integrating explainable artificial intelligence enhances transparency and supports responsible clinical adoption.
SN  - 3068-5524
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Raza2026Bridging,
  author = {Aamir Raza and Aamir Ali and Aashesh Kumar and Nikhat Fatima and Misbah Ali},
  title = {Bridging Predictive Modeling and Clinical Interpretability: An Explainable AI Approach to Parkinson’s Disease Detection},
  journal = {Biomedical Informatics and Smart Healthcare},
  year = {2026},
  volume = {2},
  number = {1},
  pages = {20-37},
  doi = {10.62762/BISH.2026.470997},
  url = {https://www.icck.org/article/abs/BISH.2026.470997},
  abstract = {Parkinson’s disease (PD) is the second most common neurodegenerative disorder worldwide, predominantly affecting older adults. Early detection is crucial, as subtle motor and non-motor symptoms frequently overlap with other conditions, often resulting in delayed diagnosis. Many existing models rely on costly and less accessible imaging modalities such as MRI or PET scans, limiting their applicability in resource-constrained settings where only routine clinical data are available. This study develops interpretable AI models for early PD detection using structured clinical variables, incorporating feature selection techniques. Feature selection was conducted via Random Forest (RF) importance ranking combined with SelectKBest statistical scoring, retaining the most informative predictors for modeling. Five classifiers were implemented in parallel: Support Vector Machine (SVM), K-Nearest Neighbors (KNN), RF, Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) network. Model performance was assessed using accuracy, precision, recall, F1-score, and ROC-AUC metrics. The RF model achieved the highest accuracy of 92.72\% (ROC-AUC: 0.968), with CNN and SVM showing competitive performance. LSTM exhibited balanced sensitivity and specificity, while KNN demonstrated relatively lower recall. To improve clinical interpretability, LIME was applied to each model to produce instance-level explanations, consistently highlighting tremor severity, motor impairment, cognitive scores, and age as key influential features. These results demonstrate that structured clinical variables alone can enable reliable PD detection without dependence on imaging. Integrating explainable artificial intelligence enhances transparency and supports responsible clinical adoption.},
  keywords = {Parkinson’s disease, explainable artificial intelligence, machine learning, deep learning, feature selection, clinical decision support},
  issn = {3068-5524},
  publisher = {Institute of Central Computation and Knowledge}
}