Research on the design of DeepMedAI Core smart medical collaboration system based on multimodal artificial intelligence

Abstract

To tackle the uneven distribution of medical resources and limited intelligent services in primary Chinese hospitals, we propose <strong>DeepMedAI Core</strong>, a multimodal innovative medical system integrating YOLOv12 for lesion detection, EfficientNet-B7 for image classification, RAG-based retrieval, and the DeepSeek-R1 large language model. The system features three core modules: medical image analysis, intelligent diagnosis, and doctor recommendation, forming a complete “data–model–service” pipeline. <strong>Results:</strong> On a brain CT benchmark, our YOLOv12 + EfficientNet-B7 pipeline achieves <strong>0.873 mAP@0.5</strong>, surpassing YOLOv12 alone by <strong>11%</strong>. The dialogue module fine-tuned on 12,842 doctor-patient dialogues, attains <strong>98.7%</strong> Top-3 diagnosis accuracy and <strong>95.2%</strong> doctor recommendation accuracy. Under 10,000-user concurrency, the system delivers <strong>99.99% availability</strong> with sub-second response latency. These results highlight the system’s effectiveness in enhancing clinical efficiency and promoting equitable healthcare through multimodal AI.

Ähnliche Arbeiten