Objective: To develop an AI training platform in PD counseling that utilizes real-time NLP processing, multimodal inputs/outputs, and adaptive reinforcement learning supported by immediate feedback in multiple aspects. We plan to make this tool accessible on GitHub to other programs and institutions for broader adoption in medical education.

Background: Effective counseling is essential for patients with Parkinson disease (PD), however inadequate training can lead to diagnostic errors(1). While patient interaction is valuable for training, neurology residents often have limited time to do so with early residency training programs mainly focused on inpatient care (2,3,4). This leads to lack of structured and safe feedback-environment (5). AI has allowed creating learning systems that can simulate patient interactions. Prior studies showed that AI-driven feedback can enhance empathy,  reporting a 20% increase in empathic responses in mental health training (6). Building on this finding, we propose a modular AI framework to improve PD counseling training through realistic, AI developed virtual patient encounters.

Method: A structured agentic AI pipeline will be implemented to integrate real-time speech recognition, sentiment analysis, and adaptive response generation (Figure 1). The system follows an autonomous perception-reasoning-action cycle with continuous learning (Supp 1) (7).  An LLM-based dialogue generates responses, vocalized through an empathic model. The FastRTC library enables low-latency audio and multimodal data exchange, while model-agnostic design ensures adaptability as AI evolves. AI will provide feedback on empathy, medical accuracy, and communication clarity. In a pilot study, 10 trainees will engage in pre/post-training PD counseling scenarios, with surveys assessing skill improvement. Voice data and feedback are collected anonymously with informed consent.

Results: Prelim Results from the baseline survey collected from 15 participants are presented in Figures 2,3 supporting the need of this AI-driven initiative.

Conclusion: This AI-driven framework is autonomous, adaptive, and intentional, suited for counseling education. It provides personalized feedback in a low-risk setting, enhancing trainee preparedness, and improving patient care while addressing language and cultural barriers. This tool could be expanded in application beyond PD counseling.

Figure 1: Agentic AI Framework

Figure 2: Preliminary Survey Results

Figure 3: Expected Benefits, Evaluation Approach

Multi-Stage Al Process for PD Counseling Training

Technical Implementation

References: 1. Newman-Toker DE, Nassery N, Schaffer AC, et alBurden of serious harms from diagnostic error in the USABMJ Quality & Safety 2024;33:109-120.
2. Yurdakul, E.S., Coskun, Z.Y., Sari, O. et al. Characteristics affecting the attitude and approach of physicians to breaking bad news: Uncertain medical situations. Humanit Soc Sci Commun 11, 490 (2024). https://doi.org/10.1057/s41599-024-02948-z
3. Li, Yaneng et al. “Leveraging Large Language Model as Simulated Patients for Clinical Education.” ArXiv abs/2404.13066 (2024): n. pag.
4. Dabir A, Arnone V, Raza B, Najib U, Pawar GV. Education Research: Appraisal of Outpatient Clinical Experience During Neurology Residency. Neurol Educ. 2023 Jan 23;2(1):e200046. doi: 10.1212/NE9.0000000000200046. PMID: 39411109; PMCID: PMC11473090.
5. Zhu J, Huang J, Cao Y, Cao L. Key challenges and countermeasures: a review of undergraduate teaching of neurology in outpatient settings. BMC Med Educ. 2025 Jan 3;25(1):19. doi: 10.1186/s12909-024-06601-w. PMID: 39754092; PMCID: PMC11699661.
6. Cao S, Fu D, Yang X, Wermter S, Liu X, Wu H. Pain recognition and pain empathy from a human-centered AI perspective. iScience. 2024 Jul 23;27(8):110570. doi: 10.1016/j.isci.2024.110570. PMID: 39211548; PMCID: PMC11357883.
7. Sandini, G., Krüger, N., & Piater, J. (Eds.). (2011). Perception-action cycle: Models, architectures, and hardware. Springer. https://doi.org/10.1007/978-1-4419-1452-1

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/multimodal-and-agentic-simulated-ai-patients-for-bridging-the-gap-in-parkinson-disease-education/