INTRODUCTION: Parkinson's Disease (PD) is a progressive neurodegenerative disorder that impairs motor and cognitive functioning, making early diagnosis and patient-specific treatment essential; however, traditional clinical assessments often delay detection and lack consistency. METHODS: This paper proposes TriFusion-PD, a multimodal Machine-Learning-Driven Targeted Nanomedicine (ML-TN) framework designed to improve early PD diagnosis and personalized therapy. The methodology integrates Transformers for analyzing temporal voice and speech patterns, Convolutional Neural Networks (CNNs) for extracting spatial biomarkers from MRI and spiral-drawing images, and Reinforcement Learning (RL) for optimizing individualized treatment recommendations. Three publicly available datasets-voice recordings, speech-signal features, and drawing/imaging data-were pre-processed, feature-engineered, and evaluated through a unified ML pipeline using metrics such as accuracy, precision, recall, F1-score, Matthews Correlation Coefficient (MCC), robustness, computational efficiency, and ROC-AUC. RESULTS: The findings show that TriFusion-PD significantly outperforms baseline models including CNN, SVM-KNN, XGBoost, LSTM, and Random Forest, achieving up to 12.5% higher accuracy, 14.2% higher precision, 15.0% higher recall, and 13.8% higher F1-score, with the highest improvement observed in MCC (16.4%), indicating more balanced and reliable PD classification across modalities. DISCUSSION: These performance gains arise from the complementary strengths of multimodal fusion, which enhances sensitivity to subtle PD symptoms, while RL-based optimization reduces misclassification and improves predictive stability under noisy or incomplete data conditions. CONCLUSION: Overall, TriFusion-PD presents a robust and clinically relevant data-driven solution for early PD detection and personalized treatment planning, demonstrating strong potential for integration into next-generation clinical decision-support systems.