CISL-PD: A deep learning framework of clinical intervention strategies for Parkinson’s disease based on directional counterfactual Dual GANs

Abstract

Listen

Translate article

Parkinson’s disease (PD) is a prevalent chronic neurodegenerative disorder characterized by both motor and non-motor symptoms. The significant heterogeneity among PD patients poses a major challenge for treatment interventions. Current clinical interventions for PD primarily target motor symptoms, often neglecting non-motor symptoms, which can lead to unnecessary complications in non-motor symptoms while treating motor symptoms. Therefore, it is crucial to provide comprehensive and precise intervention strategies that encompass both symptom types. To address this issue, we develop a deep learning framework of clinical intervention strategies for PD (CISL-PD) based on counterfactual thinking. This framework introduces Directional Counterfactual Dual Generative Adversarial Networks (DCD-GANs), which apply various counterfactual constraints to longitudinal data to generate practical and plausible counterfactual instances aligned with clinical reality. By analyzing these counterfactual instances and their differences from the original instances, we explore PD intervention strategies with duration-specific fine regulation of multidimensional features. Experiments conducted on 374 PD patients from the Parkinson’s Progression Markers Initiative (PPMI) demonstrate that the counterfactual instances generated by DCD-GANs surpass other state-of-the-art models in terms of similarity (0.307 ± 0.246), sparsity (0.513 ± 0.161), smoothness (0.238 ± 0.135), and trend consistency (0.100 ± 0.089). From these generated counterfactual instances, we develop three clinically feasible intervention strategies that address both motor and non-motor symptoms and identify corresponding patterns of PD with distinct progression differences. Validation on an independent cohort of 351 patients from the National Institute of Neurological Disorders and Stroke Parkinson’s Disease Biomarkers Program (PDBP) confirmed the framework’s robustness and generalizability. By offering precise, multidimensional intervention strategies that can address both motor and non-motor symptoms, the CISL-PD framework has the potential to enhance patient outcomes, reduce complications, improve overall quality of life, and guide clinical decision-making.