The medical large language model MedGPT is released, and AI doctors realize the transformation from effective consultation to accurate diagnosis for the first time

The medical big language model MedGPT is released, and the AI ​​doctor realizes the transformation from effective consultation to accurate diagnosis for the first time

China Times (www.chinatimes.net.cn) reporters Guo Yilin Yu Na reporting from Beijing

ChatGPT, a large-scale language model for conversational dialogue, caused a frenzy across major domestic and international media platforms in early 2023. Since then, ChatGPT has been overwhelmed by technological expansion in various fields. It is also exciting to look forward to the new diagnosis and treatment concepts that "ChatGPT Medical" will bring.

On May 25, Medical Alliance officially released its independently developed domestic first large-scale medical language model based on the Transformer architecture - MedGPT. Different from general-purpose large language model products, MedGPT mainly exerts actual diagnosis and treatment value in real medical scenarios, realizing full-process intelligent diagnosis and treatment capabilities from disease prevention, diagnosis, treatment, and rehabilitation. Leading digital medical services into the 2.0 era is because the Medical Alliance is based on MedGPT.

The reporter saw at the scene that, unlike the existing AI consultation process of 3-5 questions, MedGPT will give a preliminary suspicion after collecting enough decision-making factors through more rounds of inquiries. judge. In this regard, Wang Shirui, founder and CEO of Medical Alliance, said: "We believe that in the future, through medGPT, doctors' work and learning efficiency will be greatly improved. At the same time, high-quality medical resources will be quickly copied and promoted. Medical Alliance Internet hospitals will also use medGPT to provide more convenient, efficient and accessible medical services to patients, which is expected to further promote inclusive medical care and realize the impossible triangle of medical care."

From effective consultation to accurate diagnosis

In fact, general large language models have natural flaws in the accuracy of handling medical problems. In the consultation stage, general large language models often give conclusions easily, but for medical applications, consistency and accuracy are important. It's a bottom line issue. To ensure accuracy, MedGPT can conduct multiple rounds of interviews to collect sufficient diagnostic decisions before entering the diagnostic stage.

Wang Lei, the person in charge of the MedGPT project of the Medical Alliance, told a reporter from the China Times, "Old AI online consultation products are always very eager to give conclusions after users enter the main complaint of the case. The diagnosis is made based on only a small amount of case information. , this cannot be regarded as an effective way of consultation and communication, but more like a popular science answer."

"However, MedGPT will not give a diagnostic conclusion easily, but will guide the patient step by step for consultation, thereby giving a full picture of the condition that can support effective diagnosis. MedGPT is a human-computer interaction system with the goal of "cure". It collects sufficient information and makes medical decisions. To this end, the Medical Alliance has established a benchmarking testing mechanism for AI diagnosis and treatment accuracy and real-world doctors based on expert reviews to align AI with real diagnosis and treatment scenarios, hoping to achieve accurate Diagnosis." Wang Lei, director of the MedGPT project of the Medical Alliance, further stated.

In addition, the Medical Alliance has also established a benchmarking testing mechanism between AI diagnosis and treatment accuracy and real-world doctors based on expert reviews, constantly aligning AI with real diagnosis and treatment scenarios, and ultimately achieving accurate diagnosis.

It is reported that in order to verify the diagnostic accuracy of MedGPT, the Medical Federation extracted the files of 532 follow-up patients for information desensitization and conducted a simulated first-diagnosis experiment. The results show that the diagnostic results of Medical Alliance MedGPT are consistent with the patient’s original offline outpatient diagnosis by more than 97.5%, which fully proves the value and potential of AI doctors in the medical field.

Gives a full-process diagnosis and treatment plan

The reporter learned that based on the Transformer architecture, MedGPT can integrate multiple medical examination and detection modal capabilities, achieving a seamless connection from online consultation to medical examination for the first time. After the consultation session, MedGPT will prescribe necessary medical examination items to the patient to further clarify the condition, and the patient can be examined through multi-modal capabilities such as Medical Alliance Cloud Examination.

Based on effective consultation and medical examination data analysis, MedGPT can accurately diagnose diseases and customize treatment plans for patients. Patients can have their medicines delivered to their homes through the Medical Alliance Internet Hospital. After receiving the medicines, MedGPT will proactively provide them with medication guidance and management, intelligent follow-up reexamination, rehabilitation guidance and other intelligent disease diagnosis and treatment actions. The integration of multi-modal applications enables MedGPT to realize the entire process of prevention, diagnosis, treatment and rehabilitation.

"There are two bottlenecks in both old-fashioned AI online consultation products and online consultation products based on real doctors." Wang Lei told a reporter from China Times. On the one hand, there is no way to guarantee the accuracy of consultation. On the other hand, there is no way to move forward after the consultation to the examination. ”

In addition, Wang Lei also revealed that the MedGPT plugin application platform of the Medical Alliance has the ability to integrate more than 1,000 medical multi-modalities, which can enrich and improve the entire process of intelligent diagnosis and treatment experience. This also means that the Medical Alliance Digital Hospital has broken away from the 1.0 stage, which was based on Internet medical connection and aimed at improving efficiency; it has entered the digital medical 2.0 stage, which is based on powerful AI capabilities and realizes the intelligentization of the entire disease management process.

Using "DIAE" as methodology

As a medical professional large language model dedicated to the entire process of disease diagnosis and treatment, Medical Alliance MedGPT has formed a unique "DIAE" medical AI construction methodology, which covers Disease (disease coverage), Intelligence (intelligent), We build and polish products from four dimensions: Accuracy and Efficiency, and continuously improve the actual medical application value of MedGPT.

The Medical Alliance is accelerating the development of MedGPT to benefit more people through digital medical services. The current parameter scale of Medical Alliance MedGPT is 100B. More than 2 billion medical text data were used in the pre-training phase. 8 million high-quality structured clinical diagnosis and treatment data were used in the fine-tuning training phase, and more than 100 doctors were invested in feedback supervision. Fine-tune your training. In addition, Medical Alliance MedGPT can already cover 60% of ICD10 diseases, and will focus its research and development on multiple diseases in the near future to improve the universal coverage of digital hospitals. It is expected that by the end of 2023, it will be able to cover the medical needs of 80% of diseases.

Wang Lei calls on various types of partners in the industry, including AI technology, medicine, institutions, and medical multi-modal applications, to jointly develop and build medical application scenarios for general artificial intelligence technology and contribute to the technological development of the medical industry.

The best embodiment of new technology is to use technology to help the medical industry achieve revolutionary efficiency improvements. Currently, large-scale language model technology is developing rapidly, which will inevitably lead to major changes in the medical industry. Only through deep integration with industry can technological development fully unleash the dividends it brings. The Medical Federation will continue to promote technological development and keep moving forward to this end. It is reported that as of now, MedGPT has the first diagnosis capability for nearly 3,000 diseases, covering more than 80% of adult diseases and more than 90% of pediatric diseases aged 0-12 years old.

Editor: Yan Yuan Editor-in-Chief: Chen Yanpeng

The above is the detailed content of The medical large language model MedGPT is released, and AI doctors realize the transformation from effective consultation to accurate diagnosis for the first time. For more information, please follow other related articles on the PHP Chinese website!