A：German is a company focused on artificial intelligence imaging technology, more precisely a large vertical model company, its business focus is on the medical field of medical imaging direction. The goal is to accelerate the global medical imaging into the era of intelligence.

A：The main pain point facing the medical imaging industry is the contradiction between the huge market demand and the serious shortage of doctor resources. There are as many as 3285 medical imaging testing projects in China, and the amount of data is increasing at a rate of 30% every year. However, due to the difficulty of training doctors and the huge gap between doctors, the misdiagnosis rate and missed diagnosis rate remain high, especially in the primary medical system. The misdiagnosis rate and missed diagnosis rate may even be as high as 20% to 30%. German adopted the transformer architecture in 2017. The architecture has strong generalization understanding ability and can train a unified model -- image that can recognize all medical image images through a large amount of data, parameters and calculation power accumulation. This model has hundreds of billions of parameters, has the ability to identify all medical imaging modalities, and promotes the intelligent process in the field of medical imaging by building a proprietary model operating system and providing a localized solution that integrates training and promotion.

A：The application of small models in the field of medical imaging is subject to the underlying algorithm architecture, which requires a large amount of data annotation and poor generalization, I .e., models need to be trained separately for different types of medical images (e. g., brain CT, coronary CT, etc.), which is extremely costly. At the same time, when small models are used in pathological sections and other fields, they can only be trained for one abnormality, which cannot meet clinical needs.

A：Over the past decade, German has built two lines of business. The first line of business is intelligent equipment and systems. Through the fully automated equipment in the field of microscopic optics as a model, fool-like operations in the fields of cell morphology and chromosome karyotype analysis are realized, simplifying the operation process of the hospital. The second line of business is dedicated to combining large model technology with actual medical scenarios to provide more efficient and accurate medical services.

A：Artificial intelligence technology has brought significant changes in the medical field, especially in prenatal diagnosis. For example, amniotic fluid reporting time was reduced from 30 days to 7 days, with an accuracy rate of 99.986 percent; peripheral blood time was reduced to 4 days, and sensitivity and specificity were almost 100 percent in the field of anomaly identification. This technological advancement has lowered the threshold for industry application, making it possible to promote from prenatal diagnosis to pre-pregnancy and pre-marital populations, helping to screen out about 2% of individuals with chromosomal abnormalities in the normal population, which is of great significance to eugenics.

A：The company designated the project as "ballast stone business" because it can visually display the efficiency improvement brought about by the realization of intelligence in specific AI projects, making it easier for the public to understand the actual value of AI technology.

A：The company began to build the big model into an open platform at the end of 2024, and released a platform called MI mas in May and June 2025. The platform allows medical workers and researchers to quickly train their own exclusive models through zero-code, providing one-stop services from picture uploading, labeling tool use to model training, evaluation, deployment, etc., to help them transform medical imaging experience into curable and deployable intelligent models.

A：Based on the generalization of large models, the cost and threshold of training any exclusive project based on the underlying base model are extremely reduced, thus promoting the rapid realization of intelligence in the entire medical industry. The intelligent process includes not only training, but also deployment to daily hospitals as infrastructure, bringing explosive growth to the industry. Based on the technical advantages of the large model, if more than 3,000 projects or doctors have new diagnostic project requirements, they only need to provide 200 cases or at least 200 medical image pictures, and the training can be completed within one week at the earliest, which greatly reduces the cold start cost and the workload of repeatedly marking pictures, and greatly improves the input-output ratio.

A：The relevant documents issued by the state in November 2022 proposed that AI should become the second brain of grass-roots doctors by 2030, promote artificial intelligence medical imaging diagnosis services in hospitals at all levels, and realize the wide application of artificial intelligence technology in the medical system.

A：MI's Max platform is an operating system that has helped develop projects involving intelligent applications for more than 90 hospitals in the past ten months. By working with the team of Academician Hua Fengfeng, we have successfully completed the development of a preterm birth risk prediction model in more than 90 hospitals. The platform is not only able to quickly and accurately predict the risk of diseases (such as prostate cancer and gastric cancer), but also the cost is much lower than traditional detection methods, and the penetration rate is extremely high. In addition, the platform is still being improved, with the goal of pushing the prediction level of brain and gastric cancer to new heights, and may completely change the way of physical examination, and promote the transformation of medicine from diagnosis to accurate prediction.

A：By analyzing ultrasound data, the preterm birth risk prediction model can achieve a 0.75 AUC with only 210 cases within two weeks, with an accuracy rate of nearly 75%, surpassing the research results of tens of thousands of images in the past seven years. This model can make expectant mothers predict whether premature delivery by B- ultrasound at 16 weeks of pregnancy, and take timely intervention measures to enter the fetal protection program. When the model is combined with multi-modal data, the accuracy rate is further improved to 0.85, which exceeds the level of the director of obstetrics and gynecology with 20 years of experience. It has been used in the women's insurance system in Zhejiang Province, significantly improving the efficiency and effect of diagnosis and treatment.

A：We currently generate revenue by training AI models that serve experts and are deployed to primary hospitals and medical consortia, effectively improving the diagnostic and treatment capabilities of primary hospitals. In the future, we will continue to improve model performance, involve more experts in model training, and enable more model deployments. Our goal is to build a AI medical imaging ecology, in which excellent models will empower the primary health care system, improve the ability of primary hospitals to retain patients and refer patients to higher hospitals, thus forming a virtuous circle. At the same time, we will also make strategic adjustments based on market feedback and policy guidance to ensure that technological achievements can be effectively transformed into productivity and maximize industry and social value.

A：By 2030, the MI mas platform's case fee share revenue is expected to reach a 15.7 billion scale.

A：At present, in addition to Samsung, international giants such as GPS (General Electric, Siemens, Philips) and other imaging manufacturers are seeking to cooperate with us, hoping to empower their medical imaging equipment to complete the intelligent upgrade through our AI capabilities.

A：We hope to be a model for China's health out to sea, and plan to work with Chinese manufacturers to cooperate at the strategic level on a global scale, exploring business models from training models to model deployment, such as collecting model deployment revenue share in North America or other regions. We are currently exploring the model and are expected to see results this year.

A：We have cooperated with the top 100 hospitals in China and their experts. Some of them have purchased our all-in-one machines. For example, Zhejiang No.1 Hospital has invested more than 20 million yuan and plans to put it into 0.1 billion in the next 2 to 3 years. This shows that model training and deployment are validated, and model deployment and inference revenue will be a significant part of our revenue growth this year.

A：We are iterating on a new version of the model, which is based on 0.1 billion sheets of data and is expanding to 0.7 billion sheets of data, with a further expansion in the amount of parameters. There are two ways of deployment: for medical federations that accept online deployment, we will directly deploy them and provide API for use; For hospitals that purchase alkyd training and pushing integrated machines, we pre-install the model in the machines. At the same time, we maintain good relations with domestic chip manufacturers to assist them in model adaptation.

A：Our model is completely self-developed, the underlying architecture and every line of code is trained and developed by ourselves, and does not rely on any open source model.