DiNiu Tech: A Rising Star in Medical AI with Access to Cerebrospinal Fluid Samples from Nearly 20 Top Shanghai Hospitals

Nowadays, the influence of artificial intelligence (AI) technology is becoming increasingly profound worldwide, with its penetration into a wider range of industries. As one of the key sectors for AI integration, the healthcare industry in China has also witnessed rapid development. The issuance of the "New Generation Artificial Intelligence Development Plan" by the State Council in 2017 sent a strong positive signal to the entire health industry, making medical AI a highly sought-after area of focus.

Driven by national policies promoting industry development, an increasing number of medical AI startups are emerging. It is reported that their current business activities are primarily distributed across areas such as assisted diagnosis (including medical imaging, histopathology, and funduscopy), drug R&D, health management, hospital management, and surgical automation. Among these, 60% are mainly focused on the assisted diagnosis sector, with intelligent imaging-assisted diagnostic systems being particularly prominent, largely due to the relative maturity of image recognition technologies in the field of artificial intelligence.

Nowadays, the main track of medical AI in China seems to be saturated, with increasing homogenization. Among them, AI medical imaging is currently the most crowded sub-field, followed closely by AI pathology detection. As medical AI enters its second half, what opportunities remain for latecomers? To explore this, VCBeat interviewed Zeng Zhaopei, founder of Diuniu Technology, to discuss his views on the current development of medical AI and Diuniu Technology's strategic layout in this field.

It is estimated that there are more than 130 companies in China engaged in AI for medical imaging. The niche of pulmonary nodule screening and diagnosis alone features numerous players, including industry giants such as Alibaba and Tencent, as well as leading startups that have undergone multiple rounds of financing, such as LinkDoc Technology, Deepwise Medical, Tuma Shenwei, and Huiyi Huiying.

Why is everyone flocking to lung nodule screening? AI companies find it easy to develop products in the lung nodule sector, with short payback periods and seemingly greater potential for conversion into economic benefits. HoweverAI healthcare companies are currently burning cash, with a wide variety of algorithms whose accuracy still needs improvement, and viable payment scenarios have yet to materialize.Zeng Zhaopei, founder of Dianniu Technology, told VCBeat that under such circumstances, companies willing to devote themselves to research and development are few and far between; many merely visit hospitals with the urgent aim of marketing their products to physicians. A research article titled “Deep learning as a tool for improving the accuracy and efficiency of histopathological diagnosis,” published in recent years in the prestigious international journal Nature, shows thatThe sensitivity of the deep learning algorithm reached 100%, but the false positive rate was as high as 40%.Sensitivity refers to the probability of not missing a diagnosis in disease detection, while specificity refers to the probability of not making a false-positive diagnosis.

No matter how elegant the algorithm, it does not guarantee effectiveness in actual clinical practice. Zeng Zhaopei lamented that developing AI for healthcare is far more difficult than commonly imagined; AI in healthcare is not merely a technological field but also entails stringent professional medical requirements. Companies need to facilitate effective communication between physicians and AI technical staff during AI implementation to refine products and better serve users. The primary challenge currently facing medical artificial intelligence is that data serves as the foundation of AI, regardless of the terminal application. Furthermore, algorithmic technology alone is insufficient. The application scenarios for medical imaging are highly complex, and existing AI technologies struggle to meet clinical demands. Current industry practices focus solely on image analysis. “In the future, companies must integrate diverse patient information, such as clinical data and follow-up medical records, to create a comprehensive multimodal system.”

AI in pathology can rapidly annotate cancerous tissues, enabling pathologists to “review more cases, with greater accuracy and speed,” and clearly inform patients “what it is, whether it is cancer, and what type of cancer it is.” However, current AI-based pathological diagnostics still have limitations. The parenchymal and stromal components of solid tumors are highly complex, and the definitive diagnosis of some tumors still relies on immunohistochemistry, which undoubtedly increases the difficulty of AI-assisted diagnosis.

Furthermore, the development of pathology departments in China has been less than ideal. Due to insufficient attention from hospitals and low service fees, there has been a significant brain drain, with even some hospital pathologists producing tissue sections that fail to meet required standards. This is particularly evident inPathology departments generate low revenue and have a low level of digitalization, with a market size significantly smaller than that of medical imaging and cytological testing.With so many players entering the field of pathological AI, the intensity of future competition is easy to imagine.

Although medical artificial intelligence (AI) has been developing rapidly in China, only a small fraction of companies have successfully established viable business models. This is particularly true for the startups that have emerged in large numbers in recent years, for whom the prospects for concrete implementation of their business models remain unclear. In terms of revenue generation, medical AI employs diverse business models, such as collaborating with hospitals to provide hospital management and clinical decision support services, partnering with insurance companies to offer value-added services, and working with health checkup institutions to deliver health management and user management solutions. However, for startups, the most pressing challenge currently lies in the practical implementation of these business models.

Taking AI-powered ophthalmology as an example, disease screening services deployed at the primary care level are currently offered free of charge. Meanwhile, project operations incur costs to ensure software system updates and provide subsidies for relevant personnel. If AI systems are not deployed at the grassroots level for training, they cannot achieve continuous learning and evolution. At present, it is difficult to sell AI medical products to healthcare institutions through traditional sales models. Furthermore, AI medical products inevitably involve ongoing consumption costs. Given the current lack of a clear profitability model, no party is willing to bear this opportunity cost.

Similarly, it is difficult to price the detection of lesions by radiologists assisted by AI. Within the healthcare sector, imaging examinations have clearly defined pricing, but there is no established pricing for radiologists’ interpretation of images. How should fees be charged if the interpretation only involves identifying lesions without making judgments on their benign or malignant nature? This is an issue that medical AI companies will inevitably have to address in the future.

For startups, “survival” is the top priority. Achieving commercialization remains a widespread challenge. To continue capturing a share of this market, companies must make breakthroughs and innovate.

According to public statistics as of the end of 2018, China had 1 million medical institutions, with 95% located at the primary care level, andIn China, there are 1.1743 million inspection and testing practitioners, with a total of 6.3377 million sets of various instruments and equipment. In 2018, a total of 428 million inspection and testing reports were issued, averaging 1.17 million reports per day.China’s inspection and testing industry continues to expand in scale; however, its fundamental characteristics of being “small, fragmented, and weak” remain unchanged. Zeng Zhaopei pointed out that the role of laboratory medicine is often underestimated. In the specialized field of morphological testing, artificial intelligence requires not only advanced deep learning algorithms but also close collaboration with expert physicians. It is essential to establish diagnostic criteria for various cell types, perform precise annotation of massive cellular datasets, and conduct integrated analyses in conjunction with clinical histories. This approach enables in-depth investigation of complex and refractory cases, facilitating accurate diagnosis of numerous clinical conditions such as tumors and infections. Consequently, AI can more effectively assist physicians in their daily practice and help address real-world challenges such as the uneven distribution of medical resources.

Training algorithmic models requires a large volume of high-quality, diverse cytology images and precise, error-free expert annotations. Such data are most abundant, comprehensive, and authoritative in tertiary hospitals. However,The starting point for AI in laboratory medicine lies in Grade A tertiary hospitals, while its ultimate destination is undoubtedly the millions of grassroots healthcare institutions.Zeng Zhaopei stated that, compared with medical imaging and histopathology, the three routine laboratory tests (blood, urine, and stool) performed by clinical laboratories represent the only true opportunity to penetrate the primary care sector, as proficiency in these routine examinations is a core competency for general practitioners at primary healthcare institutions. Moreover, the Chinese government’s vigorous promotion of healthcare informatization, tiered diagnosis and treatment, and telemedicine over the past two years has created significant favorable conditions. Digital microscopy-based artificial intelligence-assisted diagnostic tools, supported by remote consultation capabilities, will empower primary hospitals across China, enabling them to effectively perform screening, triage, and diagnosis and treatment of major diseases.

Morphological examination of cells and microorganisms is a vital component of laboratory medicine. In Western medicine, it falls under the domain of cytopathology. Its clinical applications encompass routine smear examinations (including peripheral blood smears and microscopic examination of stool and urine), cytological analysis of body fluids such as pleural and ascitic fluid and cerebrospinal fluid, sputum smears, bronchoalveolar lavage fluid, exfoliative cytology of gastric juice, bone marrow smears, cervical smears, and fine-needle aspiration cytology, among others. The scope of clinical application far exceeds that of histopathology (gross pathology). It enables the diagnosis of various neoplastic and infectious diseases, serving as a critical step in determining the etiology for patients and providing an essential basis for selecting subsequent diagnostic and therapeutic strategies. According to the latest Health Statistics Yearbook, the market size of medical laboratory testing was approximately RMB 280 billion in 2018, with cell- or microorganism-related clinical tests and microbiological examinations accounting for about 30% of this total.

In 2019, the medical AI sector entered its second half, with some companies beginning to steer clear of highly saturated areas such as medical imaging, histopathology, and fundus examination, and instead expanding into other medical fields. Diuniu Technology, established in January 2019, is one such company.

Diniu Tech is a young enterprise based in Shanghai with deep roots in the medical sector. Starting with cerebrospinal fluid (CSF) cell analysis, the company focuses on the morphological AI segment within laboratory medicine. By establishing in-depth collaborations with top-tier tertiary hospitals in Shanghai and leveraging CSF specimen data from 18 hospitals, Diniu Tech is dedicated to injecting new vitality into medical AI through cellular and microbial morphological testing. The company is committed to meticulously developing a rapid and convenient AI-powered tool for clinicians.

The reasons why Diuniu Technology chose to enter the medical AI field through cerebrospinal fluid cell morphology are as follows:

First, cerebrospinal fluid (CSF) is hailed in the industry as the body’s fourth circulatory system, functioning as an independent entity. Its cellular composition is simpler than that of peripheral blood or bone marrow, facilitating rapid implementation, streamlined product development, and quicker validation of business models.

Second, cerebrospinal fluid cytological examination is an important diagnostic tool for central nervous system diseases such as infection, hemorrhage, and tumors, and it is irreplaceable.

Third, experienced cytopathologists require long-term cultivation through clinical data and have a prolonged learning curve. There is an acute shortage of cytopathologists, and even fewer are pathologists or laboratory physicians with experience in cerebrospinal fluid cytology.

Fourth, once the AI system for cerebrospinal fluid morphology is established, its application can be readily extended to peripheral blood, bone marrow, excreta smears, pleural and ascitic fluids, cervical smears, and other specimens.

Fifth, and more importantly, it benefits from unique advantages: Dianniu Technology has partnered with the cerebrospinal fluid (CSF) cytology testing laboratory of a top-tier tertiary hospital in Shanghai. Leveraging national key departments such as Neurosurgery, Neurology, and Infectious Diseases, this laboratory has established a CSF specimen bank comprising tens of thousands of cases and has gradually assumed responsibility for CSF cytology testing for dozens of tertiary hospitals and surrounding public hospitals in Shanghai.

As a distinguished graduate of Shanghai Medical College, Fudan University, founder Zeng Zhaopei previously served as a clinician at Huashan Hospital affiliated with Fudan University. He is also a seasoned entrepreneur and a rare cross-disciplinary expert bridging the medical and computer information technology sectors. With over a decade of experience in product research and development, design, and operational management within the healthcare informatics field, he initiated and founded the dermatology-focused mobile application “Skinbao” in 2015. This internet platform successfully onboarded tens of thousands of dermatologists online and attracted millions of users.

The founding team of Diuniu Technology comprises partners with extensive experience in sales and commercial monetization, as well as a team of senior professors specializing in computer vision and facial recognition. It is reported that the AI expert team consists of three renowned industry experts: Song Xiaoning, Yu Dongjun, and Feng Zhenhua. Notably, Professor Feng Zhenhua’s research on “feature point localization algorithm based on local information coupling” achieved outstanding results at top-tier global conferences in computer vision and pattern recognition, securing fourth place worldwide for profile feature point detection at CVPR 2017 and first place in the public database track of the ECCV 2018 International Visual Object Tracking Challenge (VOT 2018). Furthermore, Professor Feng was awarded the First Prize of the 2017 European Biometrics Industry Award. With only two recipients selected globally each year, Professor Feng Zhenhua was one of the two awardees that year.

It is reported that Diuniu Technology will soon launch a new AI diagnostic device in the field of cellular morphology—MicroWorld Intelligence.Zeng Zhaopei stated that, compared with symptom observation and macroscopic physical examinations, microscopic examination allows for the visualization of the “intense battle” between microorganisms and human cells, enabling physicians to identify the etiology accurately, rapidly, and conveniently. Currently, preliminary products are undergoing pilot use in select Grade A tertiary hospitals and partner hospitals in Shanghai, having garnered significant recognition and support from experts.

Next, DiNiu Tech will integrate the professional resources of this Grade-A tertiary hospital with those of other top-tier hospitals across China to establish a nationally leading Cerebrospinal Fluid (CSF) Cytology Consultation Center. The company aims to promote AI-powered morphological testing equipment to small and medium-sized hospitals, thereby enhancing their disease diagnostic capabilities. In addition, DiNiu Tech is considering initiating a financing round to intensify its R&D efforts, expanding the application of cytological morphology from CSF to peripheral blood, bone marrow, and other cell atlases, thus realizing its established business roadmap.

“During my years of entrepreneurial exploration,” Zeng Zhaopei told VCBeat, “I found the venture I am most passionate about: developing an AI-powered technological tool for physicians that enables laboratory medicine specialists, clinicians, and general practitioners alike to achieve precise diagnosis in a convenient, rapid, cost-effective, and affordable manner.”