An MLLM+RAG AI platform achieved >98% content accuracy, reduced educational video production time from 72 hours to ~10 minutes, and increased user retention rates by 40% in elderly cancer patients.
An AI-driven platform using MLLM and RAG significantly improves the efficiency of generating personalized oncology educational materials and increases engagement among elderly patients.
11075 Background: Effective patient education is critical in oncology but hindered by the complexity of treatment guidelines and the time constraints of clinicians. Traditional static materials often fail to address the personalized needs of diverse demographics, particularly the elderly. We developed an Intelligent Medical Science Popularization Platform integrating Multimodal Large Language Models (MLLM) with Retrieval-Augmented Generation (RAG) to automate the production of reliable, guideline-based oncology educational content. Methods: The platform utilizes a three-tier architecture deployed at a tertiary cancer center. The core engine combines a self-developed MLLM with RAG, anchoring content generation to authoritative sources (e.g., CSCO/NCCN guidelines and peer-reviewed journals) to ensure clinical accuracy. The system features: (1) Physician Portal: Automates the conversion of clinical protocols into patient-friendly text, layouts, and animations; (2) Multimodal Synthesis: Generates virtual avatar videos explaining diagnoses and treatments; (3) Patient Interface: Delivers personalized, accessible content via mobile apps, supporting voice interaction for elderly adherence. Results: Implementation data demonstrated significant efficiency and engagement gains. The MLLM+RAG model achieved > 98% accuracy in content generation as verified by expert oncologists. The production time for educational videos was reduced from 72 hours to ~10 minutes (> 99% reduction), enabling rapid updates aligned with new trial data. Animation production costs decreased by 10-fold. Daily content output exceeded 15 units. Notably, in a cohort of elderly cancer patients, the platform increased user retention rates by 40% and extended average session duration by 50 seconds, indicating improved engagement with health information. Conclusions: This policy-compliant AI platform validates the utility of RAG-based MLLMs in oncology. By shifting from passive information supply to active, personalized intelligent services, it significantly reduces the workload for oncologists while bridging the digital divide for vulnerable patient populations. This model offers a scalable solution for enhancing health literacy and treatment adherence in cancer care.
Liang et al. (Wed,) conducted a other in Cancer. Intelligent Medical Science Popularization Platform (MLLM with RAG) vs. Traditional manual production was evaluated on Content generation accuracy, production time, and user engagement. An MLLM+RAG AI platform achieved >98% content accuracy, reduced educational video production time from 72 hours to ~10 minutes, and increased user retention rates by 40% in elderly cancer patients.