GE Healthcare Bolsters $3B Ultrasound Business with Acquisition of AI-Powered Caption Health

On February 9, 2023, following its multi-billion-dollar acquisition of BK Medical, GE Healthcare once again made a move in the ultrasound sector by acquiring Caption Health, an AI-based ultrasound technology developer, further strengthening its ultrasound business, which generates over $3 billion in annual revenue.

Although the official parties did not disclose the specific amount of the transaction, the event itself was sufficient to send shockwaves through the industry.

On one hand, GE Healthcare’s imaging business has seen slowed growth in recent years, while its ultrasound segment continues to maintain strong momentum. A series of acquisitions, reflecting a strategy of capitalizing on this success, will effectively facilitate the deployment of ultrasound technologies in precision diagnosis and treatment as well as portable settings, thereby driving the overall advancement of the ultrasound industry.

On the other hand, as an emerging star in the field of ultrasound AI, Caption Health saw its core products, Caption Guidance and Caption AI, approved by the FDA for market launch in February and March 2020, respectively, thereby pioneering the commercialization of AI-powered ultrasound in the United States. With this acquisition by GE Healthcare, Caption Health has demonstrated its value in a distinctive manner, instilling confidence among peers in the AI healthcare sector.

Across the ocean, ultrasound AI in China is also experiencing robust growth, with nearly ten companies entering the field and covering most common clinical applications of medical ultrasound.

So, can we foster an ultrasound AI company like Caption Health? Or have we already carved out our own path in AI-powered ultrasound?

Heart failure is often referred to by physicians as the “silent killer” because its symptoms are easily overlooked, and by the time it is detected, it is often too late. However, timely ultrasound detection of early signs and early intervention can effectively slow the progression of the disease. In recent years, the number of patients with worsening heart failure in the United States has surged. Relevant data show that more than 1 million people are hospitalized annually in the U.S. due to exacerbation of heart failure, with annual expenditures on heart failure-related management reaching $40 billion.

As a U.S.-based AI ultrasound company, Caption Health closely aligns with the needs of American patients by directly focusing on the field of cardiac ultrasound. From a product pipeline perspective, both Caption Guidance and Caption AI, the ultrasound imaging software approved for Caption Health in 2020, are designed for cardiac ultrasound applications.

Among these features, Caption Guidance can analyze cardiac ultrasound examinations or two-dimensional transthoracic echocardiograms for adult patients, helping non-ultrasound specialists acquire images of diagnostic quality; Caption AI calculates the patient’s ejection fraction (used in the diagnosis of heart failure) after acquiring high-quality images.

An analysis of Caption Health’s product portfolio reveals that the company places greater emphasis on “identifying key imaging planes,” in contrast to the preference among Chinese AI ultrasound solutions for “detecting evidence of pathology and providing auxiliary diagnosis.”

Differences in product design philosophy are closely tied to national conditions. The U.S. healthcare system is more mature, with a relatively even distribution of medical resources. There, the focus is on identifying more effective data acquisition methods for specific diseases to help patients achieve early detection and timely treatment, thereby reducing health insurance expenditures. In contrast, China has a large population base and a relative shortage of physicians proficient in ultrasound diagnostics, particularly in grassroots areas. Therefore, AI in China must not only assist physicians in acquiring correct images but also support them in accurately analyzing ultrasound images.

But what kind of ultrasound-assisted products are more suitable for clinical applications in China? Sun Xuguang, Vice President and Chief Technology Officer of GE Healthcare China, believes that ultrasound examinations are more real-time than CT and MR examinations, which is more conducive to dynamic observation of images and real-time dynamic comparison. However, this examination highly depends on the doctor's experience and techniques. Different operators and different ultrasound doctors may draw different conclusions for the same patient. The empowerment of AI technology in healthcare can not only effectively supplement medical resources and help doctors shorten their learning curve, but also provide a basis for the diagnosis and treatment of more complex diseases, thereby better and faster serving patients.

However, the development of AI-powered ultrasound products presents certain challenges. First, AI is currently most commonly applied in the more mature field of radiology, where image acquisition for modalities such as CT, MRI, and DR is performed by radiologic technologists, and interpretation is conducted by radiologists. In this setting, AI processes static images and does not require real-time analysis. In contrast, ultrasound examinations generate dynamic, real-time images, with image acquisition and interpretation occurring simultaneously. Therefore, a prerequisite for achieving assisted diagnosis is the capability for “dynamic, real-time detection and diagnosis.” This effectively adds a temporal dimension to two-dimensional image processing, requiring companies not only to design multi-step processing algorithms but also to account for computational power requirements.

Secondly, the foundation of AI training is high-quality data. Due to established practices in browsing, processing, and storing ultrasound images, such data are more difficult to acquire than CT data, and exhibit a lower degree of standardization. The clarity of ultrasound images depends on the sonographer’s operational technique and the specific equipment model used. Cleaning and analyzing these data require a robust team of experts, which poses a significant challenge to corporate cost control.

Finally, the in-house development of algorithms and computing power entails additional risks. When most companies first entered the medical AI sector, they predominantly relied on open-source algorithms. However, in the field of ultrasound, there are very few reference algorithms available, forcing enterprises to develop algorithms from scratch. If the algorithms are overly complex, processing speeds will be significantly reduced; if computing power fails to meet requirements, the system will struggle to process real-time imaging data comprising dozens or even hundreds of frames. In other words, enterprises must not only bear the costs of algorithm R&D but also mitigate the risks associated with substandard performance in terms of real-time processing, dynamic adaptability, and accuracy.

Consequently, the challenges associated with AI-enhanced ultrasound have directly deterred many companies lacking adequate technical capabilities. The demand for real-time dynamic imaging in ultrasound is also reflected in algorithmic innovation; therefore, it is crucial for enterprises to possess their own algorithmic architecture that supports continuous updates and iterations. As the creator of the world’s first “real-time dynamic” ultrasound-assisted diagnostic system, MEDICAL AI has established a presence across multiple body regions, including superficial organs, abdominal organs, and cardiovascular structures. As a company with an independently owned intellectual property rights-based algorithmic framework, MEDICAL AI appears to be continually approaching and attempting to resolve the “essence of the problem.”

Technical challenges have constrained the pace of development in ultrasound AI, yet they have also established formidable technical barriers for companies that overcome these obstacles, directing them toward a market with limitless potential.

According to 2018 data from the China Association of Medical Equipment, the installed base of ultrasound systems in China was approximately 190,000 units, far exceeding that of DR (55,000 units), CT (22,000 units), endoscopes (20,000 units), and MRI (9,255 units). Considering whole-body ultrasound examinations alone, the annual number of patient visits has already reached 2 billion. With further increasing clinical recognition of ultrasound, this figure is expected to grow substantially.

Furthermore, with the continuous R&D efforts by industry leaders such as GE HealthCare and Philips Healthcare, the forms and application scenarios of ultrasound equipment are continuously diversifying.

The development of ultrasound technology has primarily followed two major trajectories. One trajectory, centered in tertiary hospitals, leverages AI empowerment to transform ultrasound from a mere imaging device into an intelligent terminal that integrates data acquisition, management, and analysis with deep learning. Its innovation lies in the synergistic integration of software and hardware.

For example, many hospitals face the issue of non-standardized patient positioning by radiologists. If the focus is solely on hardware upgrades, the positioning problem will persist regardless of how many iterations are made. To address this challenge, GE Healthcare introduced artificial intelligence and jointly launched the Senographe Crystal Nova AI Image Quality Control System for end-to-end mammography examinations.

Specifically, this AI is a new module integrated into the Nova AI Mammography System. Built upon the *Standardized Manual for Mammographic Examination*, jointly developed by experts from multiple Grade A tertiary hospitals to establish examination standards, the module strictly adheres to mammography protocols. It provides intelligent guidance throughout the entire imaging workflow, standardizing the operations of mammography technologists. While ensuring the retention of qualified images and reducing repeat scans, it also offers hospitals a traceable tool for imaging standard management. By enhancing the quality of breast cancer screening at the source of image acquisition, it empowers precise diagnosis.

Furthermore, the “Smart Ultrasound” solution within GE Healthcare’s “Edison Magic Box·Zhi” series serves as an intelligent clinical service workstation that provides AI-assisted diagnostic capabilities for multiple superficial organs during ultrasound examinations. This initiative jointly promotes the mutual recognition of examination results for the prevention and treatment of major diseases at the primary care level, thereby reducing patients’ medical expenses.

AI-powered ultrasound is equally essential in neurosurgery. For instance, ultrasound image navigation can provide precise craniotomy sites and lesion localization, aiding in surgical planning. It has become a critical tool for minimally invasive neurosurgery, particularly “keyhole” procedures. Accurate localization serves as the foundation for improving the quality of neurosurgical operations and reducing iatrogenic injury. In breast surgery, the use of ultrasound enhances the efficacy of breast-conserving surgery, significantly lowering rates of residual tumor, re-excision, and the volume of resected specimens.

Anesthesiology is a promising future application scenario for ultrasound. The use of point-of-care (POC) ultrasound in anesthesiology mainly focuses on ultrasound-guided vascular access, nerve blocks, and intraoperative transesophageal echocardiography (TEE), aiming to enhance the safety of anesthetic procedures and the accuracy of organ function assessment. This differs from diagnostic ultrasound examinations used for disease diagnosis. AI plays a role in achieving effects such as trauma assessment, image registration/fusion, system quality assurance, scanning assistance, and Doppler noise suppression.

Another major pathway is embedded within primary care settings. China’s healthcare system comprises nearly 900,000 primary care institutions. Among the three key components of medical services—clinical practice, pharmaceuticals, and diagnostics—AI ultrasound targets the “diagnostics” segment to address structural contradictions in healthcare delivery, holding promise to become a “visual stethoscope” for primary care physicians.

Unlike in tertiary hospitals, primary care facilities require not only AI to improve metrics such as “accuracy” and “efficiency,” but also a regulatory workflow for quality control of the ultrasound examination process to address issues such as incorrect patient positioning and unclear images. Furthermore, an educational module is equally indispensable; only by enhancing the service capabilities of primary care physicians can the fundamental problem of scarce medical resources at the grassroots level be resolved.

Overall, a segmented market offers greater room for the development of AI in ultrasound. Automated image quality assessment, image standardization, image segmentation, automated measurements, computer-aided diagnosis, and surgical navigation all require AI intervention to deliver value. In an interview, an ultrasound specialist from a Grade A tertiary hospital in Beijing stated that when AI empowers ultrasound by providing dynamic, real-time assistance in lesion detection and analysis, it serves as a “second pair of eyes” for sonographers. This represents a beneficial enhancement that also helps improve the ultrasound detection and diagnostic capabilities of primary healthcare institutions.

As digital empowerment becomes the norm, AI ultrasound holds significant promise for future development across all directions.

Although Caption Health was the first to generate revenue thanks to its FDA clearance, Chinese AI ultrasound companies have not fallen far behind in commercialization.

A typical example is Anke Zhen, a medical device company from Taiwan. The company has integrated its AI directly into the backend of ultrasound machines, bypassing the step of identifying valid ultrasound views and instead directly assisting physicians in analyzing ultrasound data from slices with suspicious lesions. Through this approach, Anke Zhen has achieved commercialization of its AI by entering into revenue-sharing agreements with hospitals in provinces and municipalities where its services are included in the hospital fee schedules.

Leading AI enterprise MEDICAL AI represents another model for the commercialization of AI in ultrasound. On one hand, it is crucial to achieve real-time dynamic intelligent detection and analysis of ultrasound images (including marking and analysis of abnormal areas). According to Lv Chenchong, founder and chairman of MEDICAL AI, the company’s AI-powered ultrasound product series has already been applied to multiple areas such as the breast, thyroid, and abdomen, and is currently expanding its layout to cover multiple body parts. It has now reached more than 400 medical institutions across China.

On the other hand, MEDICAL AI has established a partnership with GE Healthcare. By leveraging the device manufacturer’s mature distribution channels and extensive installed base of equipment, AI companies can achieve commercialization at relatively low costs and devote more resources to the research and development of new products. Meanwhile, medical device manufacturers can use this approach to build an AI ecosystem and accelerate the intelligent transformation of their devices. Thus, a mutually beneficial collaborative relationship is gradually taking shape.

Based on the current commercialization progress of ultrasound AI, it is not yet possible to determine which business model holds a comparative advantage. What is certain is that their path to commercialization is not as distant from that of Caption Health as one might imagine.

"Reviewing the three dimensions of application scenarios, market size, and development potential, we may already have enterprises that rival Caption Health, and even expect to build a richer AI ultrasound product matrix within a few years."

This means that both AI-powered ultrasound and conventional ultrasound are increasingly penetrating China’s healthcare system, enabling more patients to benefit from the widespread adoption of these technologies.

However, it is also important to note that while the implementation of AI in healthcare, including AI-powered ultrasound, is inevitable, it is a long-term and gradual process. Technological breakthroughs can help companies rapidly develop products, but they are unlikely to assist in educating or capturing the market.

Revisiting the entire medical AI sector, the cooling off from the rapid growth seen between 2017 and 2019 may actually be a positive development. Having moved away from the goal of disrupting healthcare, AI medical companies are now collaborating with clinicians and equipment manufacturers to establish their own pace of AI adoption. By leveraging AI-enabled ultrasound devices, they are helping to address the numerous challenges facing the healthcare system.