The integration of artificial intelligence (AI) within the healthcare industry is witnessing a remarkable surge, with influential companies such as Yidu Tech, YingTong Technology, Wanda Information, and ZhiYun Health being among the first to officially announce their partnership with DeepSeek. This vast consortium has resulted in over 30 healthcare organizations transitioning to integrate DeepSeek's advanced capabilities. These healthcare firms span multiple sectors, including drug research, imaging analysis, diagnostic screening, pathology testing, and chronic disease management, showcasing the significance of AI in transforming patient care and operational efficiency.
DeepSeek's noted attributes lie in its open-source nature and exceptional cost-effectiveness, facilitating scenarios and products previously unattainable in the healthcare domain. As one industry veteran aptly put it, while the introduction of AI heralds new possibilities, it is vital to acknowledge the hurdles regarding market validation and practical challenges like payment systems, accessibility, safety regulations, and ethical considerations that accompany these technological advancements.
One of the primary contexts where AI is finding a foothold is in the concept of "smart hospitals." A notable example is the Shenzhen People's Hospital, which recently announced its localized deployment of DeepSeek. The director of the hospital's information technology department, Ding Wanfu, explained that AI is currently employed in auxiliary diagnostics. In collaboration with Tencent, the hospital has developed an AI-powered pre-consultation service that not only prompts patients after their registration and payment but also allows doctors to generate electronic health records based on the responses provided by the patients. This integration signifies a progressive step towards digitized healthcare, enhancing both efficiency and accuracy in patient assessment.
Similarly, the Fourth People's Hospital of Shanghai disclosed that they have completed their own localized deployment of DeepSeek. Building on a repository of over 30,000 case studies and established treatment protocols, the hospital aims to provide precise decision-making support for clinicians in real-time. Moreover, on June 18, Shanghai Ruijin Hospital launched the "Ruizhi Pathology Model" in collaboration with Huawei, leveraging millisecond-level image reading capabilities to augment clinical diagnostics, thus illustrating the tactical use of AI in patient care.
In terms of diagnostic equipment, the realm of AI application diverges into a more matured domain than smart hospitals, demonstrating advancements in medical devices across myriad applications. For instance, Neusoft Medical's NeuBrainCARE software achieves an astonishing 95% accuracy in ischemic penumbra analyses, accomplishing this feat in under 90 seconds—a methodology that has gained traction within Chinese expert consensus guidelines.
Furthermore, years ago, United Imaging Healthcare incorporated AI algorithms into CT and PET-CT technologies, which enables lower radiation exposure during imaging diagnosis without compromising quality, thereby reflecting an ongoing commitment to patient safety and effective healthcare delivery.
Yet, the question remains whether these innovative applications will elevate operational efficiency and reduce costs within healthcare systems. Zhang Yuming, the head of the China Academy of Information and Communications Technology's Medical Big Data Research Center in East China, cautioned that the efficacy of these systems should be assessed through varying lenses. In contexts such as medical record writing, aiding diagnosis, and telemedicine, the AI's ability to offer transparent reasoning might enhance doctors' proficiency in evaluation and validation. Contrastingly, in scenarios involving pathological analysis and imaging navigation, AI can transcend human limits, compensating for factors like distractions stemming from human emotional states, thus underscoring the necessity for accuracy and detail in medical diagnostics.
However, as the integration of AI within medical applications flourishes, it brings forth a plethora of risks and challenges. Wang Liansheng, the head of the Artificial Intelligence Research Institute at Xiamen University, remarked that beyond universally recognized issues of data security and privacy risks, AI models face additional hurdles in areas like inference processes, accountability, and fairness. These elements raise substantial questions about the trustworthiness of the AI models, constraints tied to data input, and the need for regulatory frameworks to monitor these sophisticated AI entities effectively.
In addition to inherent business development risks, AI-powered medical devices—encompassing both software and hardware—encounter obstacles during pre-market registration and healthcare insurance acceptance stages. Zhang elaborated that AI medical devices must undergo extensive safety and efficacy evaluations and testing before they are permitted for use in patient treatments and diagnostics. Even after market approval, there remains a continuous obligation to track adverse events and analyze outcomes.
To navigate these waters, Zhang advocated for the urgent establishment of relevant standards covering myriad aspects, including industry guidelines for AI medical devices, specification for data interfaces, algorithm evaluations, and safety certifications. Therefore, fostering interoperability and ensuring sustainable health development is crucial. Moreover, ensuring the algorithms powering AI medical devices are interpretable and reliable will empower healthcare professionals and patients alike to understand the decision-making processes that underpin these technological innovations.
Regarding registration strategies for AI medical devices, Wang Jing, founder of Silicon Intelligence, proposed a meticulous classification approach. Companies ought to delineate their products accurately—those involved in diagnosing, treating, or preventing illnesses should categorically fall under the medical device classification. Compliance with pertinent regulatory requirements is essential for submitting comprehensive technical documentation and clinical data.
Furthermore, the selection of diverse and robust training and validation datasets is paramount to ensure the representativeness and diversity of the data utilized, thereby enhancing the reliability of the AI model's outcomes. The underlying algorithm's logic must also be rigorously evaluated, as it serves as a primary driver of the AI device's effectiveness.
To combat the innate challenges presented by AI "black box" phenomena, thorough clinical validation and performance assessments must underscore the AI's safety and efficacy—an approach that can lend credibility to even those algorithms that are not transparently interpretable by conventional methods.
On the topic of insurance market entry, the National Medical Insurance Administration announced on November 2024, the upcoming establishment of an "AI-assisted" extension within fields like radiological and ultrasound examinations, as well as rehabilitation projects. These initiatives empower hospitals to select between equipping their personnel for medical diagnosis or leveraging AI technologies; however, such decisions are to be made without duplicative charging at this stage. Given that various hospitals currently embed costs related to AI analysis and diagnostics within their overall service charges, a separate billing system is still relatively uncommon. AI applications predominantly function as auxiliary to human diagnostics at this moment.
Moreover, as observed by a provincial healthcare insurance official, although AI continues to serve primarily as a supportive function, technologies such as deep brain stimulation devices equipped with AI-driven natural language decoding have been in practice for years. Should such devices seek healthcare reimbursement, they may open pathways for further consideration.