Lightweight hybrid multimodal framework for Parkinson’s disease diagnosis and severity-aware chatbot-based personalized recommendations
Madankumar Nachukuru, C SushamaAbstract
BACKGROUND:
Parkinson’s disease (PD) continues to impose a substantial global healthcare burden, and early diagnosis together with individualized treatment remains challenging due to heterogeneous clinical manifestations and limited accessibility of comprehensive diagnostic tools in routine clinical practice. Existing diagnostic approaches often rely on single-modality data or handcrafted features, resulting in inconsistent diagnostic performance, limited generalization capability, and inadequate support for personalized patient management.
OBJECTIVE:
This study aims to develop a lightweight hybrid multimodal framework for accurate Parkinson’s disease diagnosis and severity-aware personalized recommendations by integrating multimodal learning with an intelligent chatbot-based clinical decision support system.
METHODS:
The proposed framework incorporates three key components. First, LightFormerNet-HW efficiently extracts discriminative deep handwriting features using a lightweight transformer-based architecture. Second, multimodal gated severity diagnosis network performs multimodal feature fusion through cross-modal attention and severity-aware gating mechanisms to enhance diagnostic accuracy and disease severity estimation. Finally, PD-GuideX, a reinforcement learning-based chatbot, generates personalized therapeutic recommendations according to the predicted disease severity, thereby supporting individualized patient care.
RESULTS:
Experimental evaluation demonstrated excellent diagnostic performance, achieving an accuracy of 98.71% and a precision of 97.63%, outperforming existing state-of-the-art approaches. The proposed framework also provides computational efficiency and improved clinical interpretability while delivering severity-specific personalized recommendations.
CONCLUSION:
The proposed lightweight hybrid multimodal framework offers an accurate, computationally efficient, and clinically interpretable solution for early Parkinson’s disease diagnosis. By integrating multimodal diagnosis with a severity-aware intelligent chatbot, the framework facilitates accessible and personalized patient management, demonstrating strong potential for deployment in real-world clinical decision support systems.