CSCO AI Aims to Transform the Full Oncology Care Cycle with a Comprehensive Tumor Knowledge Graph

China faces a shortage of high-quality medical resources, which are also unevenly distributed. For the more than 4 million newly diagnosed cancer patients each year, this means that a large number of them struggle to access appropriate medical services. The uneven distribution of medical resources poses challenges even in first-tier cities, where healthcare institutions often find it difficult to provide continuous, high-standard medical care. How can we address the scarcity of diagnostic and treatment resources? And how can we expand the coverage of standardized diagnosis and treatment protocols across China as extensively as possible? These are urgent issues that need to be resolved in the field of clinical care.

Technological advancements have provided us with an opportunity to address this pain point. As artificial intelligence continues to mature, the industry is exploring ways to leverage AI technologies to enable machines to deliver replicable services, standardize care for every patient, and achieve scalable service capacity at a sufficiently low cost. This common challenge is gradually being resolved.

CSCO AI is precisely such a solution. It was jointly developed by the Chinese Society of Clinical Oncology (CSCO) and Zhejiang Haixin Zhihui Technology Co., Ltd. (hereinafter referred to as “Haixin Zhihui”), a national high-tech enterprise. By integrating multidimensional professional knowledge—including CSCO clinical practice guidelines, experts’ clinical experience, oncology knowledge graphs, high-level clinical evidence, and adverse event management systems—it assists clinicians in formulating more standardized and precise treatment plans. Currently, Haixin Zhihui has successfully established a leading domestic intelligent oncology service platform, providing patients with one-stop, whole-course management services covering diagnosis, treatment, and rehabilitation. At the core of this achievement lies Haixin Zhihui’s industry-leading knowledge graph capability, which serves as the foundational cornerstone of the entire system.

Knowledge Graph is a concept of a knowledge network system proposed by Google in 2012. Simply put, it connects scattered information through semantic relationships and transforms it into a visualized knowledge network. Knowledge graph technology can uniformly model, organize, and manage medical data. It not only effectively describes and mines the relationships among medical knowledge but also provides strong support for higher-level medical applications, such as assisting clinical diagnosis and treatment decisions and medical question-answering systems.

With knowledge graphs as a form of knowledge representation, how to leverage them became a strategic choice facing Haixin Zhihui. Even when focusing on the medical field to build a general-purpose medical knowledge graph, the effort is largely limited to organizing and integrating knowledge and definitions related to certain diseases. “General-purpose medical knowledge graphs have significant limitations in clinical applications. They are only suitable for basic health education, patient triage, and preliminary assessments. If one attempts to delve into the specifics of disease management, clinical diagnosis, tracking, and follow-up care, various challenges arise. The unique characteristics of each disease and the expansive, interconnected knowledge system associated with clinical treatment create complexities that hinder the effectiveness of general-purpose knowledge graphs in areas such as drug selection and post-treatment rehabilitation,” said Li Yingyun, founder of Haixin Zhihui.

Therefore, Haixin Zhihui has chosen to adopt an in-depth and meticulous approach, integrating artificial intelligence to construct a knowledge graph for a whole-course tumor treatment service system. This system aims to clearly define all knowledge related to oncology, which is undoubtedly the most ideal application scenario for knowledge graphs.

“We spent several years in the early stages defining the knowledge framework spanning from cancer diagnosis and treatment initiation to comprehensive patient management, thereby establishing a relatively comprehensive oncology knowledge graph. We chose the field of oncology because its diagnostic and therapeutic system is more complex than that of other diseases, and the course of continuous treatment spans a longer period. In this context, knowledge graphs can maximize their utility in cancer care.” This is how Li Yingyun explained the rationale behind applying knowledge graphs to the oncology sector.

The second reason is that knowledge regarding the diagnosis and treatment of tumors is constantly updated every year. With technological advancements, certain diseases have been thoroughly studied, and there is no fundamental difference in treatment strategies between ten years ago and ten years later. In contrast, humanity has yet to fully understand the pathogenesis and mechanisms of tumors, making their treatment a rapidly evolving field. On a global scale, the standardization of tumor treatment in medical institutions is not particularly high. Medical professionals face significant challenges: How can they keep pace with the latest academic developments both domestically and internationally, accurately grasp the most recent clinical research progress and expansions in treatment modalities, master the newest drugs and therapeutic approaches, and apply them to the most suitable patients? “If a city has not built any new roads for 30 years, its residents do not need maps or navigation systems because nothing has changed. However, if the city is developing rapidly, undergoing ‘minor changes every three years and major transformations every five,’ even those who have lived there since childhood may need navigation guidance when traveling,” Li Yingyun used this simple analogy to illustrate the rapid evolution of tumor treatment.

Systemic anti-tumor therapies have undergone transformative changes over the past decade. While chemotherapy was the predominant modality ten years ago, clinical practice now offers a broader array of options, including targeted therapy, immunotherapy, and the latest CAR-T cell therapy, among others. This evolution underscores China’s ongoing efforts to establish a more modernized and standardized diagnosis and treatment service system, ensuring that cancer patients receive the most appropriate, regulated, standard, and suitable care.

Clinical decision support systems are the optimal tools for assisting physicians in delivering standardized treatment. The CSCO AI intelligent clinical decision support system automatically generates diagnostic and therapeutic recommendation reports after patients upload various medical diagnostic materials via a mobile app. These reports are then submitted to senior experts for review and feedback, enabling the AI model to undergo continuous closed-loop training based on the experts’ evaluations. Underpinning this operational framework is the integration of an oncology-specific knowledge graph and neural decision-support algorithms, developed by Haixin Zhihui through the consolidation of expert expertise.

Leveraging the comprehensive knowledge system of its knowledge graph, Haixin Zhihui, through CSCO AI, has not only standardized and homogenized treatment protocols in intelligent assisted diagnosis and therapy but also achieved full-course management for out-of-hospital patients, thereby creating a multi-party win-win scenario: hospitals improve overall survival rates; patients benefit from extended survival time, enhanced quality of life, and improved treatment adherence; and pharmaceutical companies, by integrating data empowerment with digital platforms, strengthen their patient support capabilities to enable precision diagnosis and treatment as well as precise rehabilitation support. Meanwhile, the clinical new drug service system based on full-disease-course management precisely matches patients with clinical trials, offering them additional therapeutic options.

"Relying on precise disease condition assessment, we establish content service capabilities covering the entire patient treatment cycle. Through multi-dimensional approaches—including building medication adherence mindsets, providing positive experience incentives, and delivering high-impact services at critical touchpoints—we create a high-value community platform. These are the distinctive features of Haixin Zhihui’s services."

Knowledge graphs serve as the foundation for clinical decision support. High-quality data and specialized knowledge systems are the cornerstones of knowledge graphs; therefore, the construction of the data layer and knowledge system is key to determining the quality of a knowledge graph.

The CSCO AI knowledge graph is not created out of thin air; rather, it effectively structures and consolidates the knowledge currently employed by clinical experts into a computable format for reuse. The most critical aspect of this process is identifying the knowledge evidence that influences clinical decision-making. Only after clearly defining this knowledge can corresponding models be designed specifically for it.

Oncology treatment is a form of evidence-based medicine, requiring robust evidence to inform clinical decision-making. However, tumors are highly diverse and specific; each type requires at least 3,000 pieces of high-level clinical evidence for therapeutic decisions. Integrating this evidence is essential to fundamentally establish the diagnostic, therapeutic, and knowledge frameworks related to oncology. “The foundational layer involves determining which patient populations, characterized by specific features, molecular subtypes, and genetic loci, should receive which treatments under what circumstances to achieve optimal outcomes,” said Li Yingyun.

After establishing the foundational level, the next step is to consider the compatibility between clinical treatment regimens and the patient’s constitution—for instance, whether the patient can tolerate the treatment, and whether underlying diseases and existing comorbidities will influence therapeutic choices. This constitutes the second level, namely, an expanded knowledge system applied in clinical practice.

In this process, Haixin Zhihui CSCO AI constructs the foundational core knowledge system for individual tumors, centered on comprehensive treatment knowledge of the disease condition, and further expands the knowledge system in clinical applications. This approach not only ensures professionalism grounded in core evidence but also demonstrates medical expertise throughout clinical practice.

Little known is the fact that “communication” poses the greatest challenge in the construction of knowledge graphs. To extract knowledge and relationships from big data, knowledge graph development requires coordinated collaboration among professionals from diverse disciplines. Therefore, applying knowledge graph technology to specific business domains is inherently a cross-disciplinary endeavor. The primary difficulty lies in determining who should integrate this cross-disciplinary knowledge. This means that builders of oncology knowledge graphs must master engineering algorithms while also understanding all terminology definitions and foundational medical knowledge related to tumors. “There is an inherent obstacle, known as the ‘knowledge barrier,’ when applying knowledge graph technology to new business domains. Those proficient in knowledge graph technology typically come from science and engineering backgrounds, whereas those with medical expertise are trained in medicine. Each group has its own established mindset, making professional interdisciplinary dialogue challenging,” stated Li Yingyun.

The second barrier lies in the understanding and organization of professional knowledge. Li Yingyun believes that a preliminary knowledge graph can only be constructed after systematically organizing such knowledge. Subsequently, business modeling is performed based on this graph, while new specialized clinical knowledge is continuously incorporated during practical applications. The third barrier concerns how computers leverage this new knowledge to conduct effective reasoning and decision-making for specific patient cases.

Therefore, the greatest challenge in constructing medical knowledge graphs lies in interdisciplinary integration, which is difficult because it demands high levels of expertise from both disciplines. This is precisely where Haixin Zhihui holds its advantage—as the artificial intelligence strategic partner of the Chinese Society of Clinical Oncology (CSCO). With the promotion and support of the Society, leading oncology experts in China, driven by a strong sense of social responsibility, devoted considerable time and effort to helping Haixin Zhihui’s technical team understand clinical pathways and clarify the underlying systemic framework. CSCO AI is the first intelligent clinical decision support product developed based on China’s clinical guidelines and practice standards. Rooted in China’s specific national context, it possesses distinct Chinese characteristics and is thus better suited to local clinical scenarios. As a result, CSCO AI has been deployed in hundreds of hospitals across dozens of provinces throughout China, benefiting tens of thousands of patients.

Moreover, the application of CSCO AI across multiple scenarios can further enhance its clinical value. For instance, it facilitates interoperability between tertiary and primary care hospitals: patients can consult online via a mobile app, while physicians leverage CSCO AI as a tool to coordinate care across different hospital tiers, thereby promoting standardized diagnosis and treatment practices at all levels.

CSCO AI can also serve as a research tool, facilitating clinical studies based on real-world cases to explore how intelligent decision-support systems assist in patient treatment and clinical applications from multiple perspectives. Furthermore, the treatment recommendations provided by CSCO AI adhere to evidence-based medicine principles, enabling its use as a quality control tool for standardized care across departments, hospitals, medical consortia, and government bodies, thereby enhancing the overall level of regional standardized treatment. Teaching hospitals may also utilize CSCO AI as a tool for case analysis learning and assessment among resident physicians. As a reference and management tool for treatment decision-making in oncology patients, CSCO AI can be integrated into various scenarios—including multidisciplinary team (MDT) discussions, physician education, ward rounds, and case conferences—thereby improving overall clinical outcomes.

Although the underlying architecture of Haixin Zhihui’s oncology knowledge graph is highly complex, it demonstrates excellent usability. For patients, following the platform’s guided prompts is all that is required, ensuring a seamless onboarding experience with no learning curve. For physicians, Haixin Zhihui’s AI-driven whole-course disease management system delivers robust capabilities in patient tracking, monitoring, follow-up, and data management throughout long-term treatment, ultimately achieving an optimal balance of efficiency among doctors, nurses, and patients.

Another highlight of this knowledge graph-based case management service is the introduction of the case manager role, which serves as a communication bridge between physicians and patients. Case managers not only assist departments in completing daily patient management tasks but also track and urge patients to adhere to treatment and rehabilitation protocols. Furthermore, they monitor and facilitate physician intervention in the management of adverse events occurring outside the hospital. With the follow-up provided by case managers, clinical management extends from in-hospital to out-of-hospital settings, thereby ensuring continuity of care.

Li Yingyun stated, “The field of oncology diagnosis and treatment will inevitably enter an era centered on patient services. With the aid of artificial intelligence, multiple stakeholders can collaborate efficiently to help patients complete the entire therapeutic management process. Of course, this system cannot be established overnight; it requires continuous evolution through practical application.”

Although Haixin Zhihui has constructed a knowledge graph with such underlying complexity, it maintains a clear understanding of the capability boundaries of both the knowledge graph and its clinical decision support system. Currently, AI-driven initiatives in areas such as pathology, medical imaging, and clinical diagnosis and treatment—including model development, quantitative analysis, feature correlation, decision support, and therapeutic efficacy prediction—serve solely to enhance physicians’ diagnostic and therapeutic efficiency, accuracy, and ability to predict treatment outcomes. These technologies empower medical practice without interfering with physicians’ authority over clinical judgment and decision-making.

Technology in Healthcare Demands Constant Reverence.