医疗大模型的生意经：一体机卖爆，医院实际使用场景却少

Core Insights - The DeepSeek model has gained significant traction in the healthcare sector, with over a thousand hospitals in China localizing its deployment within three months of its launch [1] - Initial enthusiasm for DeepSeek is waning as hospitals face challenges in practical applications, leading to concerns about its effectiveness and value [2][3] Group 1: Adoption and Initial Impact - DeepSeek has been integrated into numerous hospitals, enhancing traditional AI capabilities in areas such as pre-consultation, clinical diagnosis, imaging analysis, personalized treatment, and internal process optimization [1] - For instance, Shanghai Huashan Hospital reported a 40-minute reduction in patient wait times and a nearly 10% increase in patient satisfaction after implementing DeepSeek [1] Group 2: Challenges and Reassessment - Many hospitals are experiencing issues such as underutilization of computational resources, high error rates, and reluctance to rely on the model for clinical feedback [2][3] - A prominent hospital's IT director noted that the investment of 3 million yuan in hardware for DeepSeek did not yield the expected results [2] Group 3: Cost and Resource Allocation - The budget for localizing DeepSeek in hospitals ranges from hundreds of thousands to one million yuan, with the 70B and 32B parameter models being the most popular choices [7] - The deployment of DeepSeek requires significant hardware investment, with costs for high-performance GPUs reaching millions, which poses a barrier for many institutions [6][8] Group 4: Integration and Data Quality Issues - Hospitals face difficulties in integrating DeepSeek with existing information systems, which are often non-standardized and of varying quality, complicating data cleaning and governance [10] - The complexity of modifying existing systems raises concerns about operational risks and the potential for disruption during the transition [10] Group 5: Future Directions and Specialized Models - Some leading hospitals are exploring specialized models tailored to specific clinical needs, which have shown promising results in diagnostic accuracy [12][13] - The development of specialized models, such as those for urology and pathology, is seen as a viable path forward, potentially creating significant economic and social value in the long term [13]