Using AI to diagnose children's illnesses, today's parents are really "bold".
This spring, AI pediatricians have become one of the hot topics in the AI medical circle.
In March, Beijing Children's Hospital joined hands with Baichuan Intelligence to release the first domestic large - scale pediatric medical model and launched two versions of the "AI pediatrician" for primary - care settings and experts. In April, the Children's Hospital of Chongqing Medical University collaborated with Zuoshou Doctor to launch the "Pediatric AI Family Doctor" and a pediatric evidence - based knowledge base adapted to large - scale model applications.
Besides these two top - tier children's hospitals taking prompt action, many other hospitals are on the way to applying AI pediatric products.
In reality, the shortage of pediatricians and the difficulty for children to receive medical treatment have been long - standing problems. Since children have difficulty accurately describing their symptoms, pediatrics is also known as the "mute pediatrics". Doctors can only diagnose conditions based on limited communication and physical examinations. There are limited varieties of pediatric medications, and the dosage often depends on the doctor's discretion. Moreover, pediatrics is a "mini - general practice", and it takes at least eight to ten years to train an excellent pediatrician.
Can the "smart brain" of large - scale AI models make up for the huge gap in the demand for pediatricians?
Besides having AI serve doctors and improve the efficiency of diagnosis and treatment, can AI adjusted by doctors directly serve patients?
Is it possible for AI pediatricians to become the "killer application of medical AI"?
AI Pediatric Family Doctor: Alleviating Mothers' Concerns
After a healthy young woman becomes a mother, she often has to face many sudden medical and nursing problems, such as neonatal jaundice, intestinal flatulence, eczema, and allergies.
Out of strong concern, from major issues like developmental delays and pneumonia infections to minor ones like constipation, diarrhea, and an irregular head shape, mothers are very concerned about all their children's health problems. In the traditional Internet medical consultation scenario, young mothers are the group with a relatively high consultation frequency and a willingness to pay.
Nowadays, many users use Deepseek, Doubao, etc. to interpret laboratory reports and conduct simple consultations. However, directly applying general large - scale models to serious medical care has a high degree of hallucination, and ordinary users sometimes have difficulty judging the accuracy of the answers. Additionally, when DS and others analyze conditions and recommend prescriptions, they usually list multiple possibilities. People without professional medical training have difficulty distinguishing and choosing, posing a risk of misdiagnosis.
Therefore, developing "AI intelligent doctors" in vertical fields has become the choice of many manufacturers. Baichuan Intelligence has developed the "Baifang AI Intelligent Doctor", the official Doubao has launched the "Health Consultation Assistant" agent, and the Children's Hospital of Chongqing Medical University, in collaboration with Zuoshou Doctor, has developed the "Xiaoyi · Pediatric AI Family Doctor", etc. These 2C products are currently available for free.
So, which intelligent doctor is better in terms of consultation accuracy and user experience? A medical industry insider said that due to the great information asymmetry in medical care, it is difficult for end - consumers to quickly judge. However, there is a reference indicator: that is, how many top - tier hospitals the products developed by the AI medical manufacturer have entered, including intelligent case systems and AI auxiliary diagnosis systems.
On the one hand, hospitals have professional judgment ability; on the other hand, a product entering a hospital means "accessibility of medical data", and data is the key to training AI models.
Zhang Chao, the founder and CEO of Zuoshou Doctor, also expressed a similar view. He summarized it as: "The competitiveness of an AI doctor = the degree of access to top - tier medical resources × the thickness of personal health records." After seven or eight years of entrepreneurship, on the B - side, Zuoshou Doctor has developed products such as in - clinic translation robots and hospital information software for hospitals, serving 40 out of the top 100 hospitals.
"For the C - side, we currently focus more on becoming the 'health butler' of the public. Before treatment, users can consult the pediatric AI doctor to get judgments on the complexity of the condition, whether hospitalization is needed, and which doctor in which hospital to recommend."
He gave an example. For instance, if a baby has green stools, the AI doctor will ask the user about the feeding method, body temperature, environmental changes (such as whether the baby has caught a cold), and the evolution of symptoms (such as whether there is diarrhea or dehydration) to provide a condition analysis and judge whether immediate medical attention is needed. Then, according to the diagnosis result, a professional science - popularization video recorded by the Children's Hospital of Chongqing Medical University will be attached for the user to learn.
At present, the "Pediatric AI Family Doctors" developed by hospitals and manufacturers have also set many self - imposed constraints to control the consultation risks in serious medical scenarios. For example, they can analyze laboratory test results but do not have an entry for uploading pictures of patients' symptoms; they can recommend nursing methods but do not provide medication suggestions.
During the conversation, many interviewees mentioned that AI should learn doctors' questioning methods and thinking chains. "Medical diagnosis is a process from divergence to convergence. The consultation process needs to collect a lot of information and finally converge to a clear diagnosis," Zhang Chao believes that the past experience in developing "in - clinic translation robots" and intelligent medical records enables the AI doctor developed in cooperation with the Children's Hospital of Chongqing Medical University to understand and respond to patients' intentions relatively accurately.
Currently, this pediatric AI family doctor mainly relies on the natural traffic from the hospital side to divert offline patients to the online platform. For hospitals, the pediatric AI doctor can undertake functions such as "remote triage" before treatment and "doctor - patient education" after treatment. Moreover, in the process of spreading to a wider range of patients outside the hospital, the AI doctor can also bring patients with more treatment value to the hospital (such as bringing patients with difficult and complicated diseases to tertiary - grade A hospitals).
However, based on past experience, it is difficult for such AI medical products to obtain direct income from hospitals or medical insurance. So, how can the "pediatric AI doctor" achieve commercialization in the future?
Zhang Chao believes that a low - cost subscription fee can be charged for the 2C model, such as 1.9 yuan per day or 9.9 yuan per month. The key is to test the product's strength and user stickiness. In the future, the AI doctor can provide medical guidance for users, that is, more detailed triage and even arrange appointments. Further, the AI doctor can provide personalized health guidance for users.
Pediatrics is not a good commercialization scenario because large prescriptions (due to strict restrictions on pediatric medications) and large - scale examinations are not possible. However, the group of mothers behind pediatrics is indeed a good consumer group. They not only care about the health of their children and themselves but also influence the medical decisions of their husbands, parents, etc. "A female user from Ningxia used the AI intelligent doctor to create health records for six family members."
The above - mentioned medical industry insider believes that the AI intelligent doctor is a traffic entrance. After the user volume reaches a certain scale, there are many commercial cooperation partners, such as pharmaceuticals, hardware, and insurance. But the key is to have a certain user scale. So, who can quickly enter hospitals (acquire customers) next will become the focus of competition. "Entering a hospital is also about occupying a position. Generally, hospitals (in terms of digitalization) reject redundant construction."
AI Enters Pediatrics: Compensating for the Shortage of Doctors
As can be seen from the above, at present, AI pediatricians mainly meet the needs of "simple consultations", and more core diagnosis and treatment are still completed by B - side doctors. When generative artificial intelligence swept in, many pediatric doctors are also using AI tools to improve work efficiency and expand diagnosis and treatment ideas.
Pediatric clinical diagnosis and treatment largely rely on communication between doctors and parents and careful physical examinations of children. "I am a pediatrician with 25 years of work experience. I really like communicating with patients but don't like spending a lot of time writing and organizing medical records," said Huang Danlin, a former doctor at the Second Xiangya Hospital and now a doctor at Zhuozheng Pediatrics.
In Huang Danlin's expectation, she hopes that AI can become a doctor's assistant: before the consultation, assist in analyzing patients' health information and present past medical histories in a structured way; during the consultation, collect and record medical record information to improve the efficiency of receiving patients; help manage patients so that doctors have more time to care for patients and concentrate on core tasks such as physical examinations and handling complex decisions. "AI can be a supplement to the shortage of pediatric medical staff, rather than a replacement."
Due to the complexity of medicine, these seemingly not - difficult functions were not well - resolved before the maturity of large - language - model technology. For example, having AI automatically record the consultation process and generate a high - quality medical record, semantic understanding is the core difficulty. Traditional small natural - language models would result in a lot of redundancy and inaccuracy.
However, in the past two years, with the improvement of AI capabilities, relevant problems have been gradually solved. AI - generated medical records can not only be comprehensive and accurate but also on the basis of accurate data, some new capabilities have emerged, such as consultation association (prompting doctors what the next question should be).
This actually means that AI has moved from a simple efficiency - improvement tool to more core business and is gradually evolving from an "assistant" to a "think - tank". Dr. Gao Zheng, a pediatrician who graduated from the Shanghai Jiao Tong University School of Medicine and has been in the industry for seven years, said, "Since the emergence of ChatGPT, he has been making friends with AI tools and regarding them as professional partners and think - tanks."
Although he has not used a particularly mature "AI pediatrician" product yet, Gao Zheng believes that starting a conversation on tools like DeepSeek with the persona of "I am a pediatrician of xxx" can be regarded as using a large - language model as an AI intelligent pediatrician to some extent.
In terms of assisting in diagnostic decision - making, providing treatment ideas, and roughly estimating the course of the disease, prognosis, and risk of complications, "AI is quite powerful. The answers it gives may not be very accurate, but it is hard - working, fast, and comprehensive." Sometimes I will question its answers and ask it to provide the data sources and evidence, just like a discussion among peers."
To provide professional and accurate tools for pediatricians, Beijing Children's Hospital and Baichuan Intelligence have developed expert - level and primary - care - level AI pediatricians. In primary - care applications, the "dual - doctor" mechanism is used to improve the diagnosis and treatment level of primary - care pediatricians, such as providing auxiliary diagnosis in the early identification of children's viral encephalitis.
Due to the lack of high - quality pediatricians in primary - care settings and parents' anxiety about seeking medical treatment, the hierarchical diagnosis and treatment system in pediatrics has long been ineffective. Compared with the "hierarchical diagnosis and treatment", a professor from Peking Union Medical College Hospital once proposed the concept of "division - of - labor diagnosis and treatment": in the entire diagnosis and treatment process, besides referring patients upwards, primary - care doctors can complete some tasks. For example, with the help of tools like AI, they can conduct effective condition inquiries and physical examinations and give preliminary condition judgments.
Of course, the potential of AI is still being rapidly unleashed.
In terms of helping parents of children and saving medical manpower, "In the near future, AI can replace some manual work, such as remote triage and judging the timing of children's medical treatment," Gao Zheng believes. In addition, he predicts that in the future, with the improvement of AI's ability to "interpret and analyze videos", intelligent child development assessment and feeding guidance using AI online may also be realized soon.
"For example, if parents upload some videos of their children playing at home, AI can help analyze whether the children's behavioral development, language and social skills are normal, or conduct a preliminary screening for the risk of serious diseases such as autism and cerebral palsy. Although in the short term, AI may not be as accurate as industry experts, it should definitely be more accurate than junior and inexperienced doctors."