The hospital has connected to DeepSeek, and then what?

Sheila

follow with interest

2025-03-26

1 Comment

2134 views

1 Collection

11 minutes

  The professional skills of a product manager refer to requirements analysis, data analysis, competitor analysis, business analysis, industry analysis, product design, version management, user research, etc.

In the wave of artificial intelligence, the emergence of DeepSeek has brought new hope and challenges to the medical field. Many hospitals have joined DeepSeek in an attempt to improve their medical service levels through this advanced technology. However, the introduction of technology is not always smooth sailing, and hospitals face many problems in the actual deployment process, including high computing power costs, complex deployment mode selection, and how to ensure that the medical advice output by AI is both professional and reliable. This article will delve into the current situation and challenges faced by hospitals after integrating DeepSeek, as well as how to better serve patients with AI assistance.

In the wave of artificial intelligence, DeepSeek has emerged as a rising star, attracting global attention, and the medical field has also sparked a craze for accessing DeepSeek. Many hospitals have followed suit, attempting to improve their medical service levels with the help of this advanced technology. However, behind this hot wave of access, we need to calmly consider: hospitals have connected to DeepSeek, and then what? This is not only related to the practical application of technology, but also to the future direction of the medical industry.

1、 Application status: Seemingly blooming like a brocade

Currently, DeepSeek has shown a widespread application trend in the medical field. From the perspective of ordinary users, many people have started to rely on it to obtain health-related information, and even attempted to use it for preliminary "consultations".

There are many examples on social media where medical bloggers share experiences where patients question doctors' treatment plans based on DeepSeek's conclusions, only to discover that the medical diagnosis and treatment guidelines for the disease have been updated, and DeepSeek's answer is correct.

On the hospital side, the localization deployment of DeepSeek is progressing rapidly. Multiple hospitals, including some well-known tertiary hospitals, have completed deployment and been put into use.

After doctors input key information such as patient medical records into the system, the "AI assistant" embedded in DeepSeek can simulate the reasoning mode of professional doctors, provide detailed differential diagnosis and treatment suggestions, assist doctors in clinical decision-making, chronic disease management, medical record quality control, and remote diagnosis and treatment.

What is a B-end product manager? What is the difference between a C-end product manager and a C-end product manager?

In B2B product managers, B stands for Business, which means business. Firstly, B2B product managers need to understand the importance of this position and design product solutions that are more suitable for the project's needs. Ultimately, in their daily work, B2B product managers

View details>

The expansion of this series of application scenarios seems to show us the promising prospects of AI empowering healthcare, with a future of improved medical efficiency and more accurate diagnosis within reach.

2、 Special requirements for medical large-scale models: professionalism and reliability are paramount

The medical field has its unique professionalism and rigor, and the requirements for medical big models are much higher than those for ordinary applications. Unlike general models, medical models must emphasize causality rather than just probability based correlations.

The universal big model is indeed like a skilled puzzle player, capable of piecing together massive amounts of fragmented information to create seemingly reasonable answers, which is quite practical in daily scenarios such as searching for information and chatting.

However, medical scenarios are far more complex than ordinary puzzles - it's not just about piecing together a landscape painting, but about piecing together a precise medical map, where each piece must fit perfectly without any margin of error.

Why are general models not sufficient in the medical field?

Medicine requires' why ', not just' seems right '. The universal model can remember simple associations such as "fever+cough=flu", but it does not understand the underlying pathological mechanisms and cannot adjust judgments based on specific conditions such as the patient's allergy history.

The quality of data directly determines the reliability of the answer. If the training data contains outdated diagnostic and treatment guidelines, the model may generate seemingly reasonable incorrect answers. In the medical field, incorrect answers may directly endanger patient safety.

Lack of "after-sales verification" mechanism. The diagnosis of doctors relies on multiple layers of scrutiny such as clinical trials and expert consensus, while the output results of general models do not have such security guarantees. Once an error occurs, it is difficult to correct.

Therefore, what the medical field truly needs is a "medical customized puzzle maker" - a specialized medical big model. They must have a deep understanding of medical knowledge, use causal logic for reasoning like doctors, and be able to clearly explain "why they are fighting like this" at every step, in order to win trust in the medical scene. After all, health is not a child's play, and the puzzle must also be pieced together professionally and reliably.

In the medical field, whether it is diagnosing diseases, making treatment decisions, or communicating and interacting with patients, every step must be based on clear causal relationships.

This requires that the medical big model must master systematic and accurate medical reasoning knowledge, and its output results and interaction methods can be accurately understood, determined, and verified by medical experts. The conclusions drawn must also be able to be confirmed or falsified, that is, to achieve "interpretable" "white box".

Only by meeting these requirements can the medical big model form a consensus with medical experts, truly integrate into the professional medical system, and provide solid and reliable support for medical work. Otherwise, even if the model provides seemingly reasonable suggestions, if they cannot be explained and verified from a medical causal logic perspective, it will be difficult to gain the trust of doctors and patients, let alone be widely promoted and applied in medical practice.

3、 Challenges faced by hospital deployment: the difficult balance between cost and benefit

Despite the high enthusiasm of hospitals for accessing DeepSeek, there are many serious challenges in the actual deployment process.

Firstly, there is the issue of deployment mode. In China, due to various factors such as data security limitations, medical institutions often adopt a privatization mode for deploying AI. However, the cost of privately deployed computing power has become a burden that many hospitals cannot afford. Especially in high concurrency applications with large models, multiple GPU servers are required to form a powerful computing cluster to support it, which undoubtedly significantly increases the cost of hardware procurement, maintenance, and operation.

In order to balance practical application effectiveness with high resource consumption, hospitals must carefully and accurately select appropriate model parameters and hardware conditions based on their own application scenarios, such as outpatient volume, inpatient volume, department requirements, and possible concurrency. But the reality is that many hospitals lack sufficient professional judgment ability in this area.

With the popularity of DeepSeek, "all-in-one machines" have entered people's vision as a seemingly convenient solution. The all-in-one machine integrates and packages software and hardware, claiming to be ready to use out of the box, which is highly attractive for medical institutions with limited computing power.

But there are also problems in the all-in-one market, with huge differences in performance and price. For example, an all-in-one machine priced at over 2 million yuan can only support 3-5 users to run smoothly in actual use. For large tertiary hospitals with a daily outpatient volume of thousands, relying solely on all-in-one machines to deploy large models would require astronomical investment to meet the needs of doctors and patients.

More seriously, if medical institutions blindly follow the trend and purchase without fully evaluating their own needs and the performance of all-in-one machines, it is likely to result in the model parameters and computing power of the equipment not matching the actual needs, ultimately not only failing to achieve the original intention of improving medical services, but also causing equipment to be idle, resulting in a waste of a large amount of resources.

4、 A dose of "reassurance pill" for young doctors

In the current era of rapid development of AI technology, the anxiety of young doctors can be understood. But it must be clear that AI is still in the stage of assisting diagnosis in the medical field, and the leading role of doctors cannot be replaced.

Taking DeepSeek's integration of AI guidance as an example, in real-world scenarios, elderly patients are often more inclined to consult the guidance desk nurses directly because nurses can provide more direct and caring services, while AI guidance may encounter "bottlenecks" or provide inaccurate advice when facing complex situations.

In the emergency room, time is life. Patients may faint with just one word of 'stomach pain', and doctors need to quickly determine whether it is appendicitis or intestinal obstruction, or even more severe aortic dissection, based on experience and intuition. AI often fails to provide effective recommendations in such situations due to insufficient data. It requires complete medical history and examination data for analysis, but in the emergency room, these data are often incomplete.

In the intensive care unit (ICU), doctors face not only technical challenges but also ethical choices. And the decision will be made based on the patient's family situation, personal wishes, and the value of life. This kind of comprehensive decision-making involving ethics and morality, AI is temporarily unable to handle.

Doctors not only treat diseases, but also provide psychological support and comfort to patients. In communication with patients, doctors can convey care through language, facial expressions, and body movements, helping patients alleviate anxiety and fear. As Dr. Trudeau famously said, "Sometimes heal, often help, always comfort." This kind of humanistic care is an indispensable part of the medical process and cannot be replaced by AI.

The hospital's integration with DeepSeek is only the starting point towards "AI+healthcare", not the endpoint. At the current stage, hospitals need to have a clear understanding of the current situation and challenges of DeepSeek's application in the medical field.

Only by deeply understanding the special requirements of the medical big model, carefully dealing with the cost-effectiveness issues in the deployment process, and avoiding blindly following trends and wasting resources, can medical AI technologies such as DeepSeek truly bring positive and sustainable changes to the medical industry, improve medical quality and efficiency, and provide better medical services for patients.

This article was originally published by @ Sheila on Everyone is a Product Manager. Reproduction without the author's permission is prohibited

Image from Unsplash, based on CC0 protocol

The viewpoint of this article only represents the author himself, and everyone is a product manager. The platform only provides information storage space services

医院接入了 DeepSeek，然后呢？

Sheila关注

2025-03-26

1 评论2134 浏览1 收藏11 分钟

在人工智能浪潮中，DeepSeek的出现为医疗领域带来了新的希望和挑战。众多医院纷纷接入DeepSeek，试图借助这一先进技术提升医疗服务水平。然而，技术的引入并非一帆风顺，医院在实际部署过程中面临着诸多问题，包括高昂的算力成本、复杂的部署模式选择，以及如何确保AI输出的医疗建议既专业又可靠。本文将深入探讨医院接入DeepSeek后的现状、面临的挑战，以及如何在AI辅助下更好地服务患者。

在人工智能浪潮中，DeepSeek 异军突起，引发全球关注，医疗领域也掀起接入 DeepSeek 的热潮。众多医院纷纷跟进，试图借助这一先进技术提升医疗服务水平。然而，在这股火热的接入潮背后，我们需要冷静思考：医院接入了 DeepSeek，然后呢？这不仅关乎技术的落地应用，更涉及医疗行业的未来走向。

一、应用现状：看似繁花似锦

当下，DeepSeek 在医疗领域已展现出广泛应用的态势。从普通用户层面来看，许多人开始依赖它来获取健康相关信息，甚至尝试用它进行初步 “问诊”。

社交平台上不乏这样的例子，有医学博主分享经历，患者因 DeepSeek 的结论对医生的治疗方案提出质疑，最终却发现该疾病的医学诊疗指南已更新，DeepSeek 给出的答案是正确的。

在医院端，DeepSeek 的本地化部署进展迅速。多家医院，包括一些知名三甲医院，已完成部署并投入使用。

医生将患者病历等关键信息输入系统后，嵌入 DeepSeek 的 “AI 助理” 能够模拟专业医生的推理模式，给出详细的鉴别诊断和处理建议，辅助医生进行临床决策、慢病管理、病历质控以及远程诊疗等多项工作。

什么是B端产品经理？和C端产品经理有什么区别？

B端产品经理中的B是Business，商业的意思，B端产品经理首先就要理解这个职位的重要性，要设计出更适合这个项目需求的产品方案，B最终产品经理在日常工作中...

查看详情 >

这一系列应用场景的拓展，似乎让我们看到了 AI 赋能医疗的美好前景，医疗效率提升、诊断更加精准的未来仿佛触手可及。

二、医疗大模型的特殊要求：专业性与可靠性至上

医疗领域有其独特的专业性和严谨性，对医疗大模型的要求远高于普通应用。与通用大模型不同，医疗大模型必须强调因果性，而非仅仅基于概率的相关性。

通用大模型确实像一个擅长拼图的能手，能够将海量碎片化的信息拼凑出看似合理的答案，在日常查询资料、聊天等场景下颇为实用。

然而，医疗场景远比普通拼图复杂——它不是简单地拼一幅风景画，而是要拼一张精密的医学地图，每一块都必须严丝合缝，容不得半点误差。

为什么通用模型在医疗领域不够用？

1. 医学需要 “为什么”，而不仅仅是 “好像对”。通用模型能记住 “发烧 + 咳嗽 = 流感” 这样的简单关联，但不了解背后的病理机制，也无法根据患者的过敏史等具体情况调整判断。

2. 数据质量直接决定答案的可靠性。如果训练数据中包含过时的诊疗指南，模型可能会生成看似合理的错误答案。而在医疗领域，错误的答案可能直接危及患者安全。

3. 缺乏 “售后验证” 机制。医生的诊断依赖于临床试验、专家共识等多层把关，而通用模型的输出结果没有这样的安全保障。一旦出错，难以修正。

因此，医疗领域真正需要的是 “医学定制拼图师”—— 专用医疗大模型。它们必须深入掌握医学知识，像医生一样运用因果逻辑进行推理，并且每一步都能清晰解释 “为什么这么拼”，这样才能在医疗场景中赢得信任。毕竟，健康不是儿戏，拼图也必须拼得专业、拼得可靠。

在医疗场景中，无论是诊断疾病、制定治疗决策，还是与患者进行沟通交互，每一个环节都必须基于明确的因果关系。

这就要求医疗大模型必须掌握系统且精准的医学推理知识，其输出的结果和交互方式能够被医疗专家准确理解、确定和检验，得出的结论更要能被证实或者证伪，也就是实现 “可解释” 的 “白盒化”。

只有达成这些要求，医疗大模型才能与医疗专家形成共识，真正融入专业医疗体系，为医疗工作提供坚实可靠的支持。否则，即便模型给出看似合理的建议，若无法从医学因果逻辑上进行解释和验证，也难以获得医生和患者的信任，更无法在医疗实践中大规模推广应用。

三、医院部署面临的挑战：成本与效益的艰难平衡

尽管医院接入 DeepSeek 的热情高涨，但在实际部署过程中，面临着诸多严峻挑战。

首先是部署模式的问题，在中国，由于数据安全等多方面因素的限制，医疗机构部署 AI 多采用私有化模式。然而，私有化部署的算力成本成为了许多医院难以承受之重。尤其是在大模型高并发应用时，需要多个 GPU 服务器组成强大的算力集群来支撑，这无疑大幅增加了硬件采购、维护以及运行的成本。

为了平衡实际应用成效与高昂的资源消耗，医院必须根据自身的应用场景，如门诊量、住院患者数量、科室需求等，以及可能出现的并发数量，谨慎且精准地选择合适的模型参数和硬件条件。但现实情况是，许多医院在这方面缺乏足够的专业判断能力。

随着 DeepSeek 爆火，“一体机” 作为一种看似便捷的解决方案进入人们的视野。一体机将软硬件集成封装，号称开箱即用，对于那些自身算力建设有限的医疗机构而言极具吸引力。

但一体机市场同样存在问题，其性能和价格差异巨大。例如，一台售价 200 多万元的一体机，在实际使用中仅能同时支持 3 – 5 个用户流畅运行。对于大型三甲医院每日动辄数千人次的门诊量来说，若单纯依靠一体机来部署大模型，要满足医生和患者的使用需求，所需的资金投入将是天文数字。

更为严重的是，若医疗机构在未充分评估自身需求和一体机性能的情况下盲目跟风购买，很可能导致设备的模型参数和算力无法与实际需求匹配，最终不仅无法实现提升医疗服务的初衷，还会造成设备闲置，导致大量资源的浪费。

四、给年轻医生的一剂 “定心丸”

在 AI 技术飞速发展的当下，年轻医生的焦虑情绪可以理解。但必须明确的是，目前 AI 在医疗领域仍处于辅助诊断的阶段，医生的主导作用不可替代。

以 DeepSeek 接入 AI 导诊为例，在真实场景下，老年患者往往更倾向于直接咨询导诊台护士，因为护士能够提供更为直接、贴心的服务，而 AI 导诊在面对复杂情况时可能会出现 “卡壳” 或给出不够精准的建议。

在急诊室，时间就是生命。患者可能只说一句 “肚子疼” 就晕倒了，医生需要凭借经验和直觉迅速判断是阑尾炎还是肠梗阻，甚至可能是更严重的主动脉夹层。AI 在这种情况下往往因数据不足而无法给出有效的建议。它需要完整的病史和检查数据来分析，但在急诊室，这些数据往往不完整。

在重症监护室（ICU），医生面临的不仅是技术上的挑战，还有伦理上的抉择。并且会综合考虑患者的家庭情况、个人意愿以及生命的价值，最终做出决定。这种涉及伦理道德的综合决策，AI暂时无法胜任。

医生不仅治疗疾病，还给予患者心理上的支持和安慰。在与患者的交流中，医生可以通过语言、表情和肢体动作传递关怀，帮助患者缓解焦虑和恐惧。正如特鲁多医生的名言：“有时去治愈，常常去帮助，总是去安慰。” 这种人文关怀是医疗过程中不可或缺的一部分，也是 AI 无法替代的。

医院接入 DeepSeek 只是迈向 “AI + 医疗” 的起点，而非终点。在当前阶段，医院需要清醒地认识到 DeepSeek 在医疗领域应用的现状与挑战。

只有深入理解医疗大模型的特殊要求，谨慎应对部署过程中的成本效益问题，避免盲目跟风和资源浪费，才能让 DeepSeek 等医疗 AI 技术真正为医疗行业带来积极且可持续的变革，提升医疗质量和效率，为患者提供更优质的医疗服务。

本文由 @Sheila 原创发布于人人都是产品经理。未经作者许可，禁止转载

题图来自Unsplash，基于CC0协议

该文观点仅代表作者本人，人人都是产品经理平台仅提供信息存储空间服务