The healthcare industry is ushering in a revolution led by AI models. In the past week, a number of technology giants have joined hands with scientific research institutions, hospitals, and medical experts to launch innovative solutions aimed at improving diagnostic accuracy and optimizing medical processes.
The medical image quality control model launched by Xinchuang Haihe Laboratory was officially launched, which significantly improved the shooting quality and diagnostic efficiency of X-ray films through the powerful computing power of domestic supercomputers. Guiyang Langma Information's "39AI General Practitioner" is open for internal testing to provide efficient auxiliary diagnosis services for grassroots medical institutions. According to the latest test results of MedGPT, its AI doctors have achieved an astonishing 60% diagnostic hit rate, significantly exceeding the existing average and even higher than the real human diagnostic hit rate. The release of the "Digital Intelligence Materia Medica" model of traditional Chinese medicine has achieved the two goals of "from disease to prescription" and "from prescription to disease". China Mobile, together with the hospital and expert team, launched the first large-scale model in the field of emergency and emergency in China, which not only serves the field of first aid, but also plays an important role in the field of community medicine and telemedicine.
01
The medical image quality control model of Xinchuang Haihe Laboratory was launched
Recently, "Tianjin News" reported that the latest research and development of a large model in the medical field of Haihe Laboratory (hereinafter referred to as "Xinchuang Haihe Laboratory"), which is located in the High-tech Zone, was officially launched.
Through this software, it can quickly detect whether there are problems such as improper positioning of X-ray films during shooting, and adjust them in time to make the images clearer and avoid repeated examinations or reduce follow-up examination steps. According to reports, in response to the needs of a number of medical institutions, Xinchuang Haihe Laboratory has set up a large model research and development team, and invited imaging physicians from many hospitals in Tianjin to participate, collecting more than 8,000 standard X-ray films as sample big data, forming a set of algorithms. After more than 3 months of writing and debugging, the team has loaded the newly developed large-scale model software on the domestic supercomputer, which is expected to be officially used in clinical practice within the year. In addition, Xinchuang Haihe Laboratory is also cooperating with Chinese Insurance and other enterprises to develop large models in various professional fields such as safety production, data management, and image enhancement.
02
Langma Information's "39AI General Practitioner" is open for two closed beta versions
According to Guizhou Daily, the "39AI General Practitioner" launched by Guiyang Langma Information Technology Co., Ltd. has opened two internal test versions for ordinary users and grassroots doctors.
It is understood that the "39AI general practitioner" is a set of artificial intelligence products in the field of medical and health based on artificial intelligence large model training, which relies on the massive medical and health data accumulated by the Internet medical business of Longma Information, based on advanced natural language processing technology and deep learning algorithms, and forms an artificial intelligence "doctor's brain" through large model training methods, which will be widely used in medical personnel in primary medical and health institutions to carry out clinical auxiliary diagnosis services and provide artificial intelligence services similar to "general practitioners". Zhou Ping, vice president of Longma Information, told the media that Longma Information has cooperated with the scientific research team of Peking University Hospital to train the artificial intelligence brain for key data in the field of general medicine. In order to ensure the accuracy of the algorithm, the team members conduct regular computing power training on the "39AI General Practitioner" every month, and continuously iterate the data model to amplify the massive data of the "39AI" intelligent brain. Zhou Ping introduced that when dealing with complex cases, "39AI general practitioners" can assist doctors to sort out all the possibilities of the disease and provide treatment plans based on big data analysis.
03
A breakthrough has been achieved in the diagnosis of MedGPT by the Medical Federation
Artery.com reported that in a recent test, MedGPT had a diagnostic hit rate of up to 60%, which is significantly higher than the average diagnostic hit rate of about 35% of the existing medical model, and is also significantly higher than the real diagnostic hit rate.
Prior to this test, MediTec had conducted an internal test on MedGPT, and the test results showed that it was able to pass the complete National Medical Licensing Examination and the U.S. Medical Licensing Examination question set, and obtained a correct rate of more than 80%. However, while test results are important, they don't tell the story of real-world performance. In order to verify the true ability of MedGPT, the Medical Federation has set quite strict test conditions for MedGPT, in this test, the AI needs to further understand the patient's condition only based on medical records and through text questions and answers, and finally give a diagnostic opinion. Subsequently, the diagnostic opinion given by the AI needs to be compared with the diagnosed diagnosis to measure the ability of the AI doctor to give the correct medical diagnosis. The entire Q&A is as patient-friendly as possible to simulate a real-world situation.
- According to the test criteria, if the AI gives the same diagnostic opinion as the confirmed diagnosis, it is considered a "hit".
- If the AI's diagnosis does not contain the correct disease name, but is very similar to the target, and is judged by a professional to require further testing to confirm the diagnosis, it is considered "similar" (e.g., the actual diagnosis is "hyperthyroidism", and the diagnosis given by the AI is "thyroid disease").
Whether it is a "hit" or a "similarity", it is considered "related" to a confirmed diagnosis; Otherwise, it is considered "irrelevant". In addition to reaching a high level of 60% in the hit rate test, MedGPT's diagnostic correlation rate has increased to 84%, and the diagnostic irrelevance rate is only 16%. According to the comparative data released by Arterial Network, MedGPT even surpasses the human level.
Source: Arterial Network
04
Tasly Pharma and HUAWEI CLOUD jointly released the "Digital Intelligence Materia Medica" model
On May 9, the "Digital Intelligence Materia Medica" Chinese medicine model was officially released. The large model was jointly developed by Tasly Pharmaceutical Group and HUAWEI CLOUD on the basis of Huawei's Pangu large language model and Pangu drug molecule large model. According to reports, the large model is based on advanced tools such as Huawei's surging computing power and vector library, which can better help researchers mine and summarize the theoretical evidence of traditional Chinese medicine.
It is reported that at present, the large model collects three categories of massive data: Shouzheng (1000+ ancient books and translations, 90,000+ prescriptions, 40,000+ Chinese patent medicines, etc.), innovation (40 million + literature abstracts, 3 million + natural products, 20,000 + target gene pathway information, etc.), and industrialization (100,000+ clinical protocols, 160,000 + traditional Chinese medicine patents and pharmacopoeia policy guidelines, etc.). The large model has 38 billion parameters, based on the pre-training of massive text data of traditional Chinese medicine, combined with the retrieval enhancement of the vector library, and the fine-tuning of multiple scenarios of traditional Chinese medicine research and development, which can better help researchers complete the mining and summary of theoretical evidence of traditional Chinese medicine. The large model has three application modes: "intelligent question answering", "interactive computing" and "report generation", which can achieve the two goals of "from disease to prescription" and "from prescription to disease".
05
The first emergency first aid model - Nine Days Medical Emergency Model was launched
On May 11, China Mobile (Chengdu) Industry Research Institute, together with the Sichuan Academy of Medical Sciences, Sichuan Provincial People's Hospital, and the expert team of the emergency first aid platform of the Emergency Medicine Branch of the Chinese Medical Association, jointly released the first large model in the field of emergency first aid in China - China Mobile Nine Days Medical Emergency Model.
According to China Mobile, the Jiutian Medical Emergency Model integrates advanced AI model technology, combined with the knowledge and practical experience of emergency medicine experts, to form a unique first-aid knowledge system. Based on the full textbook data of medical majors in top universities in China and 3 million real emergency and first aid clinical data, the first aid model adopts theory + clinical strategy, selects high-quality medical data for training and optimization, and gradually "grows" into an "intelligent emergency doctor" with profound medical knowledge and rich practical experience after continuous in-depth learning. At the same time, it provides patients with more intelligent and efficient medical services by forging a number of core capabilities such as medical key information extraction, plan classification, hierarchical triage, intelligent question answering, and decision-making assistance.
According to relevant experts from the Chengdu Research Institute, with the help of China Mobile's 5G, Al, cloud network and other advantages, the Jiutian Medical Emergency Model has built a 5G emergency rescue integrated platform through innovative services such as video call, precise positioning, intelligent assistance, and on-board consultation, so as to realize the new emergency service model of "call in and positioning, call and rescue, and get into the car and hospital", helping to improve the efficiency of critical and critical treatment and speed up life treatment.
In the field of community medicine and telemedicine, the 9-day medical emergency model also plays an important role, with the help of the first aid model, grassroots medical institutions can obtain more comprehensive and professional first aid knowledge and technical support. At the same time, the large model can quickly obtain information such as the patient's condition, medical history, and treatment plan by docking with the hospital information system and medical equipment data, providing doctors with a more scientific and accurate decision-making basis.
Source | Big health school comprehensive
Join the community
Dots to share
Point Collection