Why Zhejiang Merchant Venture Chose Ultrasound as the Entry Point for AI in Healthcare: An Exclusive Interview with You Xiangdong

Whenever medical imaging AI is discussed, many readers naturally associate it with hospital radiology departments, where computer vision technology has flourished. This perception aligns with reality: according to 2021 data from the VCBeat Orange Database, there were 480 companies tagged as employing artificial intelligence technologies, among which 98 were engaged in radiology-related businesses.

However, in the broader field of medical artificial intelligence, although radiology applications were the earliest to emerge, they do not command the largest market size. As various medical AI technologies gradually mature, numerous clinical scenarios are beginning to integrate AI (“+AI”), paving the way for a surge of new AI applications.

“AI + Ultrasound” is one such example. Unlike the static images produced by CT and MRI, ultrasound generates dynamic, real-time imagery. This means that the temporal separation between image acquisition and image analysis, which is common in radiology, cannot exist in ultrasound diagnosis; this modality requires both operations to be performed simultaneously. Furthermore, unlike static images, dynamic footage is not easily archived. To conduct AI research on ultrasound imaging, it is essential to record and capture frames of specific anatomical planes at precise moments.

The high technical barriers and the lack of standardized training data have resulted in very few enterprises focusing on AI ultrasound research, with only a handful of startups venturing into this field. However, due to its potential untapped value, significant capital has already set its sights on this sector.

At the end of last year, Zheshang Venture Capital entered this sector by investing in the AI startup Shenzhi Technology. This investment firm’s first foray into medical AI helped Shenzhi Technology achieve the remarkable feat of securing three rounds of financing within a single year. Notably, the background of You Xiangdong, an investor in this round, is worth mentioning. He has previously served as Secretary to the President of Zhejiang Medical University, Head of the Preparatory Office at Sir Run Run Shaw Hospital (Zhejiang University School of Medicine), Deputy Director of the Heart Center at the Second Affiliated Hospital of Zhejiang University School of Medicine, and Vice President of the same hospital. With over 30 years of experience in clinical practice, teaching, scientific research, and management at large tertiary public hospitals, he is a seasoned clinician.

Keywords such as “first-mover advantage,” “seasoned physicians,” and “three rounds of financing in one year” have shrouded the “AI + ultrasound” sector in an air of mystery. To gain deeper insights into the unique aspects of AI-powered ultrasound, VCBeat conducted an exclusive interview with You Xiangdong, with the transcript presented below.

Q: Few companies use artificial intelligence to assist in ultrasound diagnosis. Why is this the case?

A: To answer this question, we must first establish a fundamental understanding of ultrasound. As is well known, ultrasound is one of the fastest, safest, and most cost-effective medical diagnostic tools available to licensed physicians. Whether in large tertiary hospitals or primary care settings, appropriate ultrasound equipment can be utilized to assist in patient examinations. Consequently, there is substantial demand for ultrasound examinations in China, with the annual number of scans reaching up to 2 billion. For any AI enterprise, this represents a vast market opportunity.

The demand is significant, but the shortcomings cannot be ignored. Unlike static CT images, the standardization and normalization of ultrasound diagnosis have long remained unresolved challenges, testing physicians’ experience and judgment. For example, when ten doctors diagnose the same patient, their results may vary, with some discrepancies being substantial. In other words, quality control in ultrasound examinations is extremely difficult.

This also explains why few AI companies are engaged in ultrasound. Deep learning typically requires standardized, annotated images to construct training datasets, but acquiring such data is challenging. It usually necessitates collaboration with specialist physicians, which is time-consuming, labor-intensive, and costly. In contrast, Shenzhi Technology, a company we have invested in, possesses unique industrial advantages that enable it to effectively address data-related challenges.

Q: Given that the demand existed, why did you only enter this sector in late 2020?

A: AI has indeed been a hot topic for many years, but industry hype and maturity are two distinct concepts. In recent years, medical artificial intelligence has been in a phase of exploration and discovery, with policies, technologies, and industry understanding all remaining immature. Although smart healthcare is an inevitable future trend, early-stage AI applications may be far from the final, mature implementations.

2020 marked a turning point. Within that year, nine AI-assisted medical diagnostic devices received regulatory approval for market launch, signifying both official recognition of the industry and the tangible ability of artificial intelligence to enhance hospital efficiency.

However, judging from current application scenarios, the capabilities of imaging AI have clearly not been fully leveraged. We have our own understanding of AI and hope to enter this field to facilitate its better development.

Q: To date, what problems can “AI + ultrasound” solve?

A: From a macro perspective, AI’s empowerment of ultrasound is similar to its applications in other medical specialties, primarily focusing on improving accuracy and diagnostic efficiency. However, what sets it apart is that ultrasound has a broad base of primary care scenarios as its application foundation, which endows AI-powered ultrasound with exceptional significance.

As we just discussed, ultrasound diagnosis places high demands on physicians’ experience, yet China faces a critical shortage of experienced clinicians. Since it is not feasible to bolster the supply side in the short term, can we achieve widespread access to ultrasound by lowering the barrier to its use? Shanghai Shenzhi Information Technology Co., Ltd. addresses this challenge through precisely such an approach.

Ultrasound diagnosis consists of two steps: first, acquiring correct images; second, analyzing those images. In practical applications, Shenzhi’s AI will assist physicians during examinations by identifying the optimal image segments suitable for diagnostic analysis and providing auxiliary diagnostic support. With AI support, primary-care physicians need only learn how to operate the ultrasound equipment to acquire diagnostically viable images. Once effective images are captured, physicians can either make a real-time diagnosis or upload the imaging results to the cloud for auxiliary diagnosis by more experienced specialists. In this way, AI effectively addresses two challenges: the standardization of ultrasound imaging and the lack of diagnostic experience among physicians.

However, translating vision into reality will take time. After all, village doctors cannot dedicate extended periods to learning new systems, and their examination workflows are brief, requiring product operation times to be kept within a few minutes. Therefore, it is crucial for “AI + ultrasound” solutions to achieve genuine intelligence and streamline workflows. Currently, most “AI + ultrasound” products on the market are still difficult for personnel with only nursing-level training to master quickly, indicating that further optimization of user workflows remains necessary.

Q: What is the future development potential of AI-powered ultrasound?

A: Primary care is a prominent scenario for ultrasound application, but it is by no means the only one. Within hospitals, the use of ultrasound extends beyond diagnostic services in the Department of Ultrasound to clinical departments such as Anesthesiology, Emergency Medicine, and the Intensive Care Unit (ICU). It facilitates real-time guidance and surgical assessment, enabling physicians to make more accurate evaluations, improving the efficiency of diagnosis and treatment for life-threatening critical conditions, and allowing for more effective monitoring and regulation of basic vital functions as well as protection and support of vital organs.

Taking the Department of Anesthesiology as an example, the application of point-of-care (POC) ultrasound in this specialty primarily focuses on ultrasound-guided vascular access, nerve blocks, and intraoperative transesophageal echocardiography (TEE), with the aim of enhancing the safety of anesthetic procedures and the accuracy of organ function assessment. This differs from diagnostic ultrasound used for disease diagnosis. The role of AI in this context includes facilitating trauma assessment, image registration/fusion, system quality assurance, scanning assistance, and Doppler noise suppression.

Of course, for large-scale medical equipment, AI currently serves more as a complementary enhancement; however, it is foreseeable that its role will become increasingly critical in the future.

Q: How should we view the business model of AI in healthcare?

A: The business model of AI is a well-worn topic, yet it remains the greatest challenge facing the survival of artificial intelligence companies to this day.

When investing in AI, we considered three questions: Has AI improved the accuracy of diagnosis and treatment? Has it enhanced the efficiency of doctors and nurses, thereby boosting both the social and economic benefits of hospital patient care? Is there a viable commercial pathway? From the current perspective, the answers to the first two questions are affirmative, but for the third, we can only seek clues through experience.

The journey of an AI product from design to profitability involves multiple steps, with regulatory approval being a critical milestone. Currently, no AI-enhanced ultrasound products have received approval from the National Medical Products Administration (NMPA). However, given the approvals granted to auxiliary diagnostic software for fractional flow reserve (FFR), pulmonary nodules, and diabetic retinopathy, such approval is likely imminent.

Next is identifying the payer, which is both the most challenging and the most critical aspect. Generally speaking, end-users (B2C) are unlikely to become payers. The logic of asking patients to pay extra for AI-assisted diagnosis during a hospital visit is impractical, as patients clearly place greater trust in their physicians than in an intangible algorithm.

Will hospitals become payers? This depends on whether AI delivers economic benefits to them. Heartflow, a U.S.-based company specializing in fractional flow reserve (FFR) technology, offers a compelling model. In the management of coronary artery disease, it is often difficult to determine through conventional imaging alone whether invasive angiography or stent placement is necessary, leading to routine use of coronary angiography for nearly all patients. To address this, Heartflow developed FFRCT, a non-invasive tool that physiologically assesses the severity of coronary stenosis. By reducing unnecessary coronary angiograms, FFRCT helps hospitals lower healthcare costs. Consequently, this emerging technology has been rapidly adopted by hospitals.

To date, the market for AI-powered ultrasound remains predominantly focused on primary healthcare. If the technology can truly lower barriers to entry, enabling more primary care physicians to perform diagnostics and thereby reducing patients’ disease-related expenditures over the long term, its social value would be immeasurable.