Modern Medical AI: Shattering Conceptual Illusions
A significant portion of human life, nearly one-third, is spent in slumber. Yet, an alarming number of individuals are grappling with sleep disorders, particularly obstructive sleep apnea (OSA). Recent statistics indicate that in 2023, around 200 million people aged 30 to 69 in China are afflicted with OSA, making it the country with the highest prevalence of this condition in the world. Surprisingly, the diagnosis rate for OSA in China is less than 1%, contrasting sharply with the approximately 20% diagnosis rate observed in the United States. This disparity suggests a vast population of undiagnosed OSA patients in China, highlighting a pressing need for improved diagnostic methods and treatments.
Accurate diagnosis is fundamental for patients suffering from OSA to receive the necessary treatment and care. Currently, polysomnography is often regarded as the "gold standard" for diagnosing sleep disorders. However, it comes with several drawbacks: patients are required to undergo monitoring in a laboratory setting under the supervision of trained professionals, which can be expensive and uncomfortable. The procedure involves attaching multiple electrodes to the scalp and face, which can be invasive and lead to poor patient tolerance.
Advertisement
Addressing these core diagnostic challenges, there has been a surge of interest in artificial intelligence (AI) technologies exploring innovative solutions. Recently, a major event was held: the finals of the "JD Health Global Medical AI Innovation Competition," co-hosted by the China Artificial Intelligence Association (CAAI) and JD Health. Under the theme of "intelligent algorithms for sleep monitoring," competing teams showcased their OSA detection algorithm models, promising to lower the barriers for diagnosing sleep disorders.
This competition represents the first medical AI contest in the internet healthcare industry, with JD Health as a prominent player committed to advancing its capabilities while simultaneously offering a platform to spotlight undiscovered innovations. It aims to establish a foundation for the practical application of medical AI technology.
The medical AI sector in China took its first steps relatively late, embarking on research in this field only in the 1980s. Nevertheless, due to the rapid evolution of AI technology, its applications within healthcare have become more widespread, offering solid assistance to doctors through integrated diagnosis and treatment. The impressive data processing capabilities of AI have notably enhanced the efficiency and accuracy of disease diagnoses and continues to improve patient experiences in the healthcare system.
Nonetheless, the development of medical AI hasn’t been without obstacles. The serious nature of the healthcare industry complicates the application of AI compared to other sectors, raising the bar for the already scarce AI talent in healthcare. Compounding these issues is the prevalent phenomenon of data silos within the medical industry, which results in low coherence and integration across the industry chain. This makes it challenging for AI technology, which relies heavily on vast amounts of data, to progress, thus slowing the deployment of concrete applications.
In light of the established trends and challenges in medical AI development, JD Health launched the "JD Health Global Medical AI Innovation Competition" in June of this year with a keen eye on AI innovations. The competition saw a remarkable turnout with 1556 participants forming 1284 teams, resulting in more than 10,000 submissions. Following extensive rounds of preliminary eliminations, semi-finals, and code evaluations over three months, 20 teams advanced to the finals, where their work was rigorously assessed by industry experts leading to the crowning of award winners.
Utilizing its substantial background in AI innovation and insight into industry needs, JD Health set two core themes for the competition: "intelligent algorithms for sleep monitoring" and "innovative applications of large medical models."
Under the latter theme, competing teams introduced outstanding contributions, such as the electronic medical record quality control and generation platform based on a multi-modal large model, the FEET foot monitoring system, and a long-term diabetes management model named LLMAgent, alongside a deep learning-based skin cancer identification system.
The "AI Knight" startup team proposed a "vision-based sleep monitoring algorithm," which classifies original physiological signals while transforming them into image formats for classification. This innovative approach culminated in a multi-modal model that accurately identifies apnea events while maintaining a compact model size suitable for various wearable devices.
In the realm of large medical models, the "Language School" team presented a platform that adeptly processes and comprehends various forms of data—audio, video, images, and text—to automatically generate outpatient records, inpatient records, and discharge summaries while ensuring accuracy and safety through pre-processing filters and post-control engines. This boosts the efficiency of medical professionals tremendously.
Both the AI Knight and Language School teams secured first prizes in their respective themes, showcasing the intersection of AI technology and healthcare. The significance of this competition lies not only in presenting innovative applications but also in highlighting the necessity of collaboration among skilled professionals to drive deeper integration between healthcare and AI.
The finals of the "JD Health Global Medical AI Innovation Competition" embodied an exciting convergence of innovations in the healthcare sector, with many curious and brilliant minds dedicated to overcoming modern medical challenges.
In 2023, with the resonance of large model technologies echoing across multiple industries, JD Health introduced a medical large model, "Jingyi Qianxun," leveraging the capabilities of JD Group's "Lingxi" general model. This has greatly accelerated the comprehensive AI deployment of its products and solutions, forming a robust matrix aimed at consumers, medical experts, and healthcare initiatives.
Unlike many generic models that excel in data capabilities but lack specific application avenues, "Jingyi Qianxun" has been incredibly active since its launch. According to Wang Guoxin, head of the Intelligent Algorithm Department in JD Health’s technology product sector, "Jingyi Qianxun" has focused on revitalizing previous applications, especially in areas that assist physicians such as consultation support systems and patient management.
As explained by Wang, prior to the rise of large models, JD Health had already implemented a system designed to assist doctors. For example, in its 2020 prospectus, it outlined how it had utilized various AI applications to optimize pre-consultation inquiries, prescription renewals, and medication management, thereby enhancing the precision and efficiency of medical tasks.
From its early stages as an AI technology facilitator to becoming a leader in medical large models, JD Health has demonstrated a consistent dedication to investing in and exploring AI technologies.
In July of this year, MedBench updated its evaluation rankings for Chinese medical large language models. "Jingyi Qianxun" achieved an impressive composite score of 92.4, placing it at the top of the leaderboard.
As the model’s abilities receive authoritative recognition, JD Health is also expanding its range of application scenarios. In April 2022, the company launched China's first skin-specialty Internet hospital, providing premier telemedicine services for dermatological conditions. The AI-assisted diagnosis accuracy at this facility exceeds 95%, and the conversion rate for follow-up service payments has reached 20% among skin disease patients.
At the 2024 China International Service Trade Fair in September, JD Health's skin specialty hospital was recognized as the only internet medical representative and awarded the "Innovative Service Demonstration Case," further highlighting its role as a leading practitioner in medical AI.
In terms of implementing large models, JD Health’s psychological services underwent a third upgrade, unveiling an AI companion named "Chat Recovery Universe," which features customizable roles that enable empathetic interactions during dialogues. The company has also ventured into consumer-side products with a health assistant named "Kangkang," aimed at efficiently addressing users’ health inquiries while accurately matching them to healthcare resources.
Within the heart of the healthcare landscape lies a hospital setting where JD Health continues to deepen its strategic collaboration with Wenzhou Medical University First Affiliated Hospital. Their jointly developed "pre-consultation digital doctor based on large language models" has been selected as a typical case study for emerging technology innovation applications at the 2024 China Hospital Information Network Conference (CHIMA2024).
With abundant intrinsic scenario developments driving the implementation and iteration of large models, and continuous engagement with external ecosystems injecting resources, JD Health’s journey showcases the role of AI in propelling the intelligent development of healthcare services. However, the medical sector, having one foot in the AI era, is still faced with the challenge of disparate levels of technological penetration across various segments. The complete transition to AI readiness is yet to be achieved.
The groundwork for the advancement of medical AI has been laid at the policy level. In March 2023, the directive to foster the "Internet+ healthcare" strategy was introduced to facilitate the integration of the Internet and AI in health services.
In 2022, the National Health Commission proclaimed the "14th Five-Year" plan for digital health information, emphasizing the need to enhance the standards for new-generation information technologies, including "Internet+ healthcare" and medical AI. This supports a variety of applications such as emergency care, remote consultations, distance examinations, clinical auxiliary diagnosis, public health services, and hospital management.
As large model technologies find their footing in the medical health arena, corresponding standardized systems are also being developed rapidly. Recently, the China Academy of Information and Communications Technology announced the establishment of a standard system for large models in the healthcare industry. This infrastructure includes essential technical requirements, testing methodologies, and maturity assessment series.
According to a report by Global Market Insights, the "AI+ healthcare" market is projected to grow at a compound annual growth rate exceeding 29%. Experts estimate that China’s "AI+ healthcare" market will surpass a 30% compound annual growth rate over the next ten years.
Despite favorable policy support and a burgeoning market landscape, significant challenges still confront the wider industry, including medical data quality, model development issues, and social dynamics. Issues such as low-quality data diminishing AI model precision, scarcity of data surrounding rare diseases leading to uneven data distribution, excessive imbalance in data distributions potentially skewing algorithms, and the "black box" issue regarding AI, along with patients' acceptance and trust in AI technology, remain prominent barriers.
Participants across the medical AI field must not only meet specific demand scenarios but also evolve their technological capacities to overcome these challenges. This effort cannot be accomplished by any one enterprise alone; it requires synergistic collaboration between smart medical equipment manufacturers, large data model firms, AI diagnostic service platforms, and healthcare organizations opening their doors to provide real-world scenarios while academia injects fresh perspectives and talents into the industry.
The Global Medical AI Innovation Competition serves as a prime example of deep integration among industry, academia, and healthcare. The finals featured judges from prestigious institutions like CAAI, Beijing University of Posts and Telecommunications, Beijing University of Technology, Tsinghua University, and Beijing Institute of Technology, alongside industry professionals from JD Health, all conducting evaluations of the competing teams.
According to JD Health CEO, Jin Enlin, the competition has created an open, collaborative platform for domestic AI innovators, sparking creativity and intelligence among practitioners that can address current challenges and pain points in the healthcare sector. He hopes more individuals will join this promising field, and JD Health is willing to co-create greater value for society and the industry with everyone involved.
Pioneering the medical AI landscape, JD Health understands that maximizing value requires joint progression across the industry, leading to the inception of this groundbreaking competition. This initiative serves as a window showcasing that medical AI is pivotal in transforming and upgrading China’s healthcare sector and that the integration of various stakeholders from industry, academia, and healthcare represents the robust foundation for accelerating the development of medical AI.