BingAI Medical Secures Breakthrough in Ovarian Cancer Diagnosis with Low-Cost, High-Accuracy AI Model Based on Routine Lab Tests

Among the cancer threats facing women, ovarian cancer has quietly become a major concern for women's health due to its insidious onset and lack of typical clinical manifestations. Most patients are unfortunately already in the middle or late stages at the time of diagnosis, with a poor prognosis; its high mortality rate ranks first among malignant tumors of the female reproductive system.

Therefore, early screening has also become a critical line of defense in safeguarding women's health: by detecting subtle signs of precancerous lesions, physicians can promptly identify patients with early-stage ovarian cancer and rapidly initiate intervention or treatment strategies.

However,However, Professor Zhen Xin from the School of Biomedical Engineering at Southern Medical University revealed to Chengguo Bureau that the ovarian cancer markers currently widely used in clinical practice—carbohydrate antigen 125 (CA125) and human epididymis protein 4 (HE4)—have limitations in both sensitivity and specificity, making it difficult to meet the urgent need for precise diagnosis.

“Whether from the perspective of in-depth academic analysis or in response to the urgent demands of clinical practice, we aim to break through existing bottlenecks and explore and establish new ovarian cancer biomarkers or diagnostic methods.”Based on this, Professor Zhen Xin, in collaboration with a team of physicians, has developed a low-cost, easily accessible artificial intelligence-assisted diagnostic tool with high accuracy, facilitating precise prevention and control as well as early diagnosis of ovarian cancer.

More than two years ago, Professor Liu Jihong from Sun Yat-sen University Cancer Center, Professor Qinglei Gao from Tongji Hospital affiliated with Tongji Medical College of Huazhong University of Science and Technology, and Professor Li Xiao along with Professor Xin Zhen’s team from the Women’s Hospital School of Medicine Zhejiang University engaged in in-depth discussions and literature reviews. They established a multidisciplinary research collaboration model integrating medicine and engineering, gradually initiating studies on the development of artificial intelligence-assisted models for ovarian cancer diagnosis.

Zhen Xin stated that the unique feature of this artificial intelligence diagnostic model lies inIntegrating Laboratory Test Indicators. The so-called laboratory test indicators refer to routine tests such as complete blood count, biochemical assays, coagulation analysis, and urinalysis; they are not only the cornerstone of clinical diagnosis and treatment but also an indispensable part of routine health examinations.

“Laboratory test indicators have been standardized in clinical practice, with relatively low costs and widespread adoption across various health checkup programs and healthcare institutions at all levels..” Zhen Xin explained, “If they can be used as screening markers for ovarian cancer, this will undoubtedly greatly promote the early detection of the disease. This is especially valuable in primary healthcare institutions and routine physical examinations, holding immeasurable value for improving the diagnostic level of ovarian cancer, strengthening secondary prevention strategies, and enhancing patient prognosis.”

To this end, Zhen Xin’s team first collected over 10,000 cases of data from three hospitals: Sun Yat-sen University Cancer Center, Tongji Hospital Affiliated to Tongji Medical College of Huazhong University of Science and Technology, and Women’s Hospital School of Medicine Zhejiang University. This dataset included patients with ovarian cancer, benign adnexal lesions, and normal health check-ups, covering 98 laboratory test results and clinical characteristics. Subsequently, the team integrated artificial intelligence technology to develop an ovarian cancer prediction model—the MCF model—which ultimately identified 51 laboratory indicators and age as core variables.

In an interview, Zhen Xin stated that the development of the model involved a rigorous process of data cleaning, feature selection, and performance evaluation. The final set of 51 laboratory indicators encompasses complete blood count, biochemical tests, stool analysis, and urinalysis, which, when integrated with age factors, form a comprehensive, multidimensional predictive system. Through both internal validation and independent external validation, the team demonstrated that the MCF model not only performs exceptionally well on the internal dataset (with an AUC of approximately 0.949) but also remains robust in external validation, highlighting its excellent generalizability and promising application prospects.

It is worth mentioning that, the MCF model significantly outperforms traditional biomarkers CA125, HE4, and their combined detection in identifying ovarian cancer patients, particularly those with early-stage disease, in terms of both AUC and sensitivity.“Even in the absence of tumor marker data such as CA125, the MCF model can still accurately predict ovarian cancer risk by leveraging remaining routine laboratory tests and age information,” said Zhen Xin.

For doctors and patients, this is a more comprehensive and user-friendly diagnosis and treatment platform.

On one hand, it achieves the comprehensive utilization of routine laboratory test indicators.Traditional methods often rely on only a few indicators for modeling, whereas this model comprehensively integrates all collected experimental metrics. By conducting comprehensive analysis and leveraging their respective strengths while mitigating weaknesses, it enhances the accuracy of prediction results.

On the other hand, this model demonstrates strong adaptability and applicability.“It can process various types of complex clinical data, including data missing due to patients not completing all tests. ‘Even when faced with such imperfect data, the model can still provide relatively accurate predictions. This strong adaptability to real-world clinical data makes the model highly valuable in practical applications,’ said Zhen Xin.”

However, during the R&D process, Zhen Xin’s team also encountered several difficulties and challenges. First, the complexity and diversity of clinical data posed significant hurdles to model development. Zhen Xin stated, “Real-world data often suffer from flaws and missing values, which necessitate meticulous data preprocessing and feature selection during the modeling process to ensure the accuracy and stability of the model.”

Secondly, the heterogeneity of multi-center data is another challenge they need to overcome. Data from different hospitals may vary in terms of quality, distribution, and other aspects, making it difficult to directly apply models developed at one hospital to others. Furthermore, model interpretability and transparency are also critical aspects that the team must address.

To address the aforementioned difficulties and challenges, Zhen Xin’s team implemented a variety of measures and approaches. For instance, during the data preprocessing phase, they employed multiple techniques to clean and organize the data; in the model construction phase, they utilized advanced machine learning algorithms and feature selection techniques to enhance the accuracy and stability of the models; and during the model validation phase, they conducted multiple rounds of internal and external validation to ensure the generalizability of the models...

As the project advances, the Zhen Xin team also plans to further expand the platform’s functionalities, providing clinicians with reference support for treatment selection and prognostic assessment in ovarian cancer.“We aim to provide a comprehensive and meticulous solution that covers every critical link from early screening to treatment and even prognosis. During the treatment phase, we focus on accurately predicting patients’ therapeutic responses, such as anticipating drug resistance in advance and forecasting potential complications or adverse reactions associated with the treatment. Furthermore, we extend our attention to post-treatment recurrence risk assessment, tailoring personalized early warning and response strategies for each patient,” said Zhen Xin.

In June 2023, Zhen Xin met Chen Yiqun, a partner at Shangjun Investment, at a salon event. At that time, the team’s research project was nearing completion and preparing for journal submission, while the preliminary results had inspired Zhen Xin to consider commercializing the technology.

From an industry perspective, Chen Yiqun told VCBeat that the concept of AI-assisted diagnosis is gaining momentum, and there is already a consensus in the medical community on the critical importance of early screening and early diagnosis. However, how to reduce costs, improve diagnostic efficiency, prevent the excessive concentration and waste of medical resources, and promote the widespread adoption of technology remain significant challenges for technical teams. The low-cost, accessible, and highly accurate AI-assisted diagnostic tool developed by Professor Zhen Xin is precisely what clinical practice needs.

Thus,In October 2023, with the support of Shangjun Investment, Zhen Xin established Bingaike Medical to further advance product registration and industrialization.

Zhen Xin cited two reasons for partnering with Shangjun Investment: first, the successful translation of research findings requires close coordination across all stages, and Shangjun Investment is an institution specializing in the interdisciplinary integration of medicine and engineering, which aligns closely with the team’s needs; second, its geographic proximity greatly facilitates communication and collaboration between the two teams. “Whenever we encounter challenges, we receive prompt professional guidance and advice,” said Zhen Xin.

Transitioning from a professor to an entrepreneur, Zhen Xin is well aware of the significant differences between academic paradigms and business models. “In my view, companies focused on translating research outcomes must adhere to their unique underlying logic to achieve sustainable, long-term growth. Academic thinking emphasizes depth and breadth, whereas business thinking prioritizes efficiency and market returns. Although my background is primarily in scientific research, I recognize that in the business world, the right approach is to let professionals handle their respective areas of expertise.”Therefore, Zhen Xin believes that his primary focus should be on scientific research to ensure the project’s technical robustness and reliability, while leaving commercial operations to more specialized teams.。