Shukun Multimodal Medical AI Model: Bringing the 'Digital Doctor' to Life

On the banks of the Huangpu River, a grand event has commenced. From November 15 to 16, 2024, the 31st National Congress of Radiology of the Chinese Medical Association (CCR2024) was grandly held at the Shanghai World Expo Center. Themed “Standardization and Innovation, Leading Development,” this congress aimed to promote multidisciplinary integration and innovative development in medical imaging.

Aligning with the conference theme, SHUKUN presented four major solutions—Intelligent Imaging, Intelligent Clinical Care, Sci-Tech Innovation Platform, and Multimodal Medical-Specific Large Language Model—demonstrating its steadfast progression from imaging to clinical practice. Notably, SHUKUN’s self-developed “Shukun Kun” multimodal medical-specific large language model showcased cutting-edge AI technologies and genuine capabilities in assisting clinical diagnosis, drawing crowds to SHUKUN’s booth and making it the focal point of the event.

As a significant branch of the AI industry, medical AI has witnessed rapid growth in recent years, driven by three converging forces: the swift advancement of artificial intelligence technologies, favorable national policies, and the digital economy’s emergence as a key driver of global economic growth. The development of AI technology has also spurred the explosive growth of large-scale pre-trained models. Particularly in the healthcare sector, the application of large models in clinical diagnosis and treatment, disease prediction, and health management helps promote the industry’s digital transformation and meet the public’s health needs. It is precisely for these reasons that numerous medical large models have emerged overnight like mushrooms after rain.

Although large medical models are experiencing robust growth and enjoying a broad market, they impose stringent requirements on data collection and quality. Their deep learning training capabilities remain to be fully validated, diagnostic accuracy is still insufficient, and significant challenges must be overcome before they can be truly integrated into clinical practice.

Long before the emergence of ChatGPT, SHUKUN had already recognized the immense value of large language models and strategically positioned itself in advance. After years of research and development, SHUKUN has launched its multimodal medical large model, “Shukun Kun.” This model boasts “five core capabilities”: multimodal data understanding, deep learning, imaging diagnosis, diagnostic report generation, and clinical reasoning. These five capabilities establish “Shukun Kun” as a truly clinically viable medical large model.

Shukun’s multimodal large medical model is capable of multimodal understanding and processing across video, text, semantics, and medical imaging. It establishes comprehensive connections among imaging data, clinical practice, and patients, thereby intelligently providing physicians with real-time knowledge support and decision-making assistance. This enables “Shukun Kun” to assist human doctors in efficiently completing diagnostic and therapeutic tasks with expert-level reasoning comparable to that of mid-to-senior-level physicians, substantially improving diagnostic accuracy and healthcare efficiency, and truly realizing “human-AI collaboration.” Such capabilities are rare in the industry. Moreover, Shukun’s multimodal large medical model has successfully withstood rigorous evaluation by human experts.

On the leaderboard released by CMB, the largest Chinese medical evaluation benchmark, “Shukun Kun” achieved the top rank with state-of-the-art (SOTA) performance. Practice has demonstrated that the deep learning capabilities of SHUKUN’s multimodal large medical model far surpass those of large models developed by international tech giants in both the knowledge-based and case-based tests of this medical licensing examination.

At the Beijing Industry Large Model Innovation Application Competition, hosted by the Beijing Municipal Science and Technology Commission, SHUKUN’s multimodal medical large model won the First Prize, ranking first in the medical field. During the competition, the model achieved a 98% accuracy rate in single-disease diagnosis and over 80% accuracy in diagnosing complex cases with multiple comorbidities. Furthermore, in subsequent clinical demonstration validations, it helped double physicians’ diagnostic efficiency, earning acclaim from the expert judges present.

As SHUKUN’s multimodal medical large model made its debut, sweeping multiple awards and having its capabilities validated by the market, some may ask: With numerous large models already in the market, why has SHUKUN been able to develop a top-tier large model?

The answer is simple: it stems from long-term accumulation leading to breakthrough success.

In the field of medical imaging, offerings range from Coronary CTA AI, the world’s first artificial intelligence product for cardiovascular imaging, to CTA+CT FFR, which has impressed the industry by enabling non-invasive assessment of both coronary vessel “morphology and function”; from Head and Neck CTA AI, which saves patients in critical, life-threatening situations, to Whole-Chest CT AI, equipped with hundreds of functional parameters to achieve “zero” missed diagnoses of ground-glass nodules; from Fracture AI, which automatically detects bone injuries and diseases to alleviate emergency department pressure, to non-gated calcium scoring for non-invasive, precise early screening of coronary heart disease; as well as Liver MR AI, a globally unique solution that automatically detects lesions and performs quantitative and qualitative analysis.

In clinical practice, CT-FFR enables non-invasive assessment of coronary hemodynamics, accurately diagnosing coronary stenosis with functional impairment, thereby reducing unnecessary coronary angiography and revascularization procedures. Plaque analysis AI achieves over 90% accuracy in identifying calcified and non-calcified plaques, precisely assessing the degree of luminal stenosis and predicting acute coronary events, thus providing strong support for clinical decision-making. The one-stop stroke solution rapidly determines stroke type, identifies the culprit vessel, performs post-processing vascular analysis, quantifies the affected area, and facilitates swift decision-making for thrombolysis and thrombectomy. Preoperative lung planning AI quickly generates 3D reconstructions of pulmonary vasculature, anticipates anatomical variations in advance, and provides precise, personalized surgical plans, helping physicians perform surgeries safely and efficiently while minimizing surgical risks and postoperative complications.

On one hand, SHUKUN has accumulated substantial expertise and resources in the fields of medical imaging and clinical practice. On the other hand, SHUKUN’s team has consistently adhered to its original mission of “creating a dedicated digital doctor for every individual.” This commitment provides the essential foundation and support for the continuous evolution of large language models, enabling them to progress steadily and sustainably on this new journey. Ultimately, this allows the technology to be effectively deployed in commercial scenarios, truly empowering both the industry and end users.

With its cross-modal and interdisciplinary capabilities spanning video, images, text, and audio, the SHUKUN large language model can collaborate with human physicians to deliver medical services efficiently, employing clinical reasoning comparable to that of mid-to-senior-level specialist physicians.

With a diagnostic accuracy rate of up to 98% for single diseases and an 80% concordance rate in complex cases involving multiple comorbidities, the SHUKUN large language model demonstrates the capability to assist physicians at varying levels of expertise, thereby enhancing the standardization and consistency of clinical diagnosis and treatment.

Thanks to its deep learning capabilities, it can assist physicians in rapidly and standardizing the completion of medical record documentation. This means that SHUKUN’s large language model can ensure that diagnostic conclusions made by junior physicians and those practicing at primary care institutions are accurate and comprehensive, with no errors or omissions.

Trained on a vast volume of clinical cases and data, and equipped with expertise in medicine and engineering as well as proficiency in various data analysis tools, the SHUKUN large language model can accumulate valuable data, provide research ideas to physicians, and generate charts and models, thereby facilitating the implementation and publication of scientific research.

Because it can interpret patient case data, imaging data, and biochemical results and conduct in-depth analysis, the SHUKUN large model enables personalized diagnosis and provides one-on-one treatment plans;

Unconstrained by time or space, whether in home settings, community hospitals, or more complex treatment scenarios, this means that SHUKUN’s large language model can deliver patient-centered, full-lifecycle health management, acting like a caring and dedicated family physician to provide patients with compassion and care.

Perhaps in the future, SHUKUN’s large models will be capable of even more. What is easy to foresee, however, is that SHUKUN’s multimodal medical large model will be able to provide services to patients and healthcare institutions at every level, truly bringing “digital doctors” to everyone’s side.

This is SHUKUN’s vision, and the ultimate destination for its multimodal medical large language model. When that day arrives, the healthcare industry will usher in its own era of boundless possibilities, and everyone will be able to enjoy a healthy life empowered by artificial intelligence.