Personalized rTMS treatment recommendation with retrieval-augmented LLM reasoning.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Clinical Psychiatry

Background:

• Repetitive transcranial magnetic stimulation (rTMS) is a key therapy for major depressive disorder (MDD) and treatment-resistant depression (TRD).
• Current rTMS protocols often lack personalization, failing to address individual patient variability.
• There is a need for data-driven approaches to optimize rTMS treatment parameters.

Purpose of the Study:

• To develop and evaluate a novel, interpretable framework for personalized rTMS treatment recommendations.
• To integrate patient profiles and clinical data for customized rTMS protocol generation.
• To enhance the efficacy of rTMS therapy through individualized treatment strategies.

Main Methods:

• A retrieval-augmented generation (RAG) framework combining sentence embedding models and large language models (LLMs).
• Patient profiles are encoded into semantic representations for retrieving similar clinical cases.
• LLMs use retrieved cases for few-shot, in-context learning to synthesize personalized rTMS parameters (frequency, intensity, mode).

Main Results:

• The framework achieved a high rTMS protocol matching accuracy of 78.18% using Bge-large-en-v1.5 for retrieval and GPT-4o-mini for reasoning with 15 few-shot examples.
• The approach successfully integrates multiple rTMS parameters for comprehensive personalization.
• The system demonstrated interpretability and fine-tuning-free operation.

Conclusions:

• The proposed AI framework offers a promising, data-driven approach to personalized rTMS therapy for depression.
• This method enhances treatment customization by considering individual patient characteristics.
• The framework is adaptable for resource-poor clinical settings, advancing neurostimulation therapy.