FUAS-Agents: Autonomous Multi-Modal LLM Agents for Treatment Planning in Focused Ultrasound Ablation Surgery

Abstract

Focused Ultrasound Ablation Surgery (FUAS) has emerged as a promising non-invasive therapeutic modality, valued for its safety and precision. Nevertheless, its clinical implementation entails intricate tasks such as multimodal image interpretation, personalized dose planning, and real-time intraoperative decision-making processes that demand intelligent assistance to improve efficiency and reliability. We introduce FUAS-Agents, an autonomous agent system that leverages the multimodal understanding and tool-using capabilities of large language models (LLMs). The system was developed using a large-scale, multicenter, multimodal clinical dataset of over 3000 cases from three medical institutions. By integrating patient profiles and MRI data, FUAS-Agents orchestrates a suite of specialized medical AI tools, including segmentation, treatment dose prediction, and clinical guideline retrieval, to generate personalized treatment plans comprising MRI image, dose parameters, and therapeutic strategies. The system also incorporates an internal quality control and reflection mechanism, ensuring consistency and robustness of the outputs. We evaluate the system in a uterine fibroid treatment scenario. Human assessment by four senior FUAS experts indicates that 82.5\%, 82.5\%, 87.5\%, and 97.5\% of the generated plans were rated 4 or above (on a 5-point scale) in terms of completeness, accuracy, fluency, and clinical compliance, respectively. In addition, we have conducted ablation studies to systematically examine the contribution of each component to the overall performance. These results demonstrate the potential of LLM-driven agents in enhancing decision-making across complex clinical workflows, and exemplify a translational paradigm that combines general-purpose models with specialized expert systems to solve practical challenges in vertical healthcare domains.

Figures (14)

• Figure 1: Overview of FUASAgents’ data flow and data format. The framework illustrates the end-to-end data flow and agent collaboration for FUAS treatment planning. Patient-specific inputs, including MRI imaging reports, basic clinical information, and raw MRI images imported from the hospital HIS system, are first processed by the Planner Agent, which coordinates task decomposition and workflow scheduling. MRI images are encoded and passed to the Executor Agent for image segmentation and dose prediction, producing lesion annotations, feature vectors, and stratified dose estimates. Based on these intermediate outputs, the Strategy Agent synthesizes treatment analysis and personalized planning strategies in textual form. An Optimizer Agent subsequently performs autonomous quality assessment and optimization using rule-based clinical constraints and predefined principles. The final output is a structured treatment planning report integrating imaging results, dose recommendations, and treatment strategy.
• Figure 2: Performance evaluation and feature analysis of the dose prediction model. (A) Receiver operating characteristic (ROC) curve of the combined model, illustrating its discriminative performance; (B) Decision curve analysis (DCA) of the combined model, demonstrating the net clinical benefit across different threshold probabilities; (C) Importance ranking of key radiomics features in the radiomics-based model; (D) Statistical significance of clinical predictors in the clinical prediction model.
• Figure 3: Ablation Study Results.For more details, please see Appendix C.
• Figure 4: Human evaluation across different models
• Figure 5: Four representative uterine fibroid cases