Returning to the Essence of Healthcare: How AI Solves Real Clinical Problems at Scale

On March 11, as the Fourth Session of the 13th National People’s Congress drew to a close, China’s “Two Sessions” neared their end.

As the “Two Sessions” convened during the inaugural year of China’s 14th Five-Year Plan, marking the first year of the country’s entry into a new stage of development, healthcare has remained a top priority for public welfare. VCBeat has compiled selected policy recommendations and proposals from participating deputies and committee members concerning the healthcare sector. Overall, key areas of focus include strengthening primary healthcare, promoting the inheritance and development of traditional Chinese medicine (TCM), fostering pharmaceutical innovation, expanding basic medical insurance coverage while supplementing it with commercial health insurance, encouraging and supporting the domestic substitution of medical devices and pharmaceuticals, and advancing the application of digital health technologies.

Among the numerous healthcare-related proposals, the widespread advocacy for AI technology merits discussion. This technology has played a crucial role in areas such as empowering primary care and advancing precision medicine research. Currently, while many companies are seeking application scenarios for artificial intelligence across various medical institutions, few have managed to identify the genuine, pressing needs for this technology from the perspective of “returning to the essence of healthcare and solving clinical problems,” thereby achieving large-scale implementation. To address this challenge, we must analyze the three stages of medical AI: demand exploration, large-scale deployment, and commercialization.

Whether it is the coordinated reform of medical care, health insurance, and pharmaceuticals, or the tiered diagnosis and treatment system, a major reason for the impediments to policy implementation lies in the lack of infrastructure in primary healthcare—namely, weak informatization foundations and severely limited physician resources in terms of both capacity and quantity. These factors further erode trust in primary healthcare services, leading to underutilization of relevant resources and failure to fulfill their intended role.

Tao Xiaodong, General Manager of the Smart Healthcare Division at iFlytek, stated: “Due to the limited capabilities of primary healthcare institutions, residents generally do not choose them as their first point of care when falling ill. This is because patients are unable to determine whether their condition is mild or severe. In particular, urban residents tend to visit large hospitals even for common symptoms. Consequently, primary care facilities experience low outpatient volumes and have underutilized medical resources. As a result, many common symptoms and diseases that should be managed at the primary level are instead treated at large hospitals, where specialists end up handling routine cases. This constitutes both an underutilization of primary healthcare resources and a waste of high-quality medical resources in large hospitals, representing a ‘double waste’ within the healthcare system.”

Popularizing AI in primary care can address this issue. Specifically, by leveraging the clinical decision support system (CDSS) capabilities of AI, primary care settings can “create” a virtual assistant equipped with general practice knowledge to aid physicians in diagnosis and treatment.

Guided by this approach, BAT, iFlytek, and major health IT enterprises have developed corresponding general practice AI+CDSS solutions to enhance the level of informatization in primary healthcare and improve diagnostic accuracy at the grassroots level, aiming to fundamentally address challenges in primary care.

It is worth noting that while Clinical Decision Support Systems (CDSS) can indeed address issues such as resource shortages within individual institutions, they do not resolve the problem of medical data standardization. In 2019, a municipal Health Commission in Guangxi published the document “New Measures, Existing Problems, and Recommendations for Implementing the Tripartite Coordination of Medical Care, Health Insurance, and Pharmaceutical Services,” which pointed out that one of the difficulties in executing this tripartite coordination was the incomplete implementation of “coordination.” Specifically, without a comprehensive database containing complete patient information, regulatory authorities are unable to effectively supervise and manage the healthcare institutions involved in patients’ medical-seeking behaviors.

To address this issue, Iflytek has identified an effective solution. Specifically, the company collaborates with regional health commissions to establish standards and promote the implementation of Clinical Decision Support Systems (CDSS) within these regions, thereby resolving interoperability challenges among primary healthcare institutions and between primary and higher-level medical facilities.

As of February 28, 2021, the AI Medical Assistant, designed to address primary healthcare needs, had been deployed in over 30,000 primary healthcare institutions across more than 170 districts and counties nationwide. It served over 50,000 primary care physicians, benefited more than 50 million residents, provided over 400,000 auxiliary diagnostic recommendations daily, and delivered more than 120 million AI-assisted diagnoses cumulatively. This has effectively enhanced the quality of diagnosis and treatment at the primary care level, setting a benchmark for the large-scale implementation of AI in the industry.

The needs of tertiary hospitals differ significantly from those of primary care institutions; the latter prioritize standardization and consistency, whereas the former require a profound understanding of medicine to address challenges encountered in clinical practice.

Take the field of medical imaging as an example. Throughout the entire diagnosis and treatment workflow, medical imaging serves as a critical tool—but not the sole one—for physicians in diagnosing patients. Factors such as age, past medical history, complete blood count (CBC), and even pathological biopsy results all constitute evidence supporting final clinical decisions. Therefore, AI capabilities should be aligned with the specific needs of radiologists and clinicians.

Tao Xiaodong, General Manager of the Smart Healthcare Division at iFlytek, told VCBeat: “Radiology AI is currently a relatively mature subfield. While AI can indeed improve the accuracy and efficiency of physicians’ diagnoses, it does not ‘replace’ them. During prolonged image interpretation, physicians may overlook certain potential imaging details due to fatigue; in such cases, algorithm-based AI can help identify these oversights, thereby enhancing diagnostic completeness and physician efficiency. Furthermore, since AI is influenced by algorithms and data, it provides consistent judgments for given images, unlike humans whose assessments may vary due to environmental factors or personal state. In other words, AI offers a high degree of consistency.”

Based on the above analysis, we can clearly identify the application scenarios for AI in medical imaging—there is indeed a demand for its use in radiology departments, clinical departments, and high-volume third-party imaging centers. This demand stems from the need to “liberate workforce capacity.”

AI voice entry is a more mature AI application technology in tertiary hospitals, with its earliest implementation dating back six years.

In hospital dental departments, clinicians often have both hands occupied during diagnosis and treatment, making it impossible to document medical records in real time. Consequently, they either rely on memory to complete the documentation after the consultation or enlist a nurse to assist with record-keeping.

The drawback of the former is that non-real-time medical records are prone to overlooking details, significantly increasing the likelihood of errors; the drawback of the latter is the additional expenditure on human resources, which will lead to increased treatment costs for patients and hospitals.

Addressing this need, the Chinese AI-listed company iFlytek has developed a specialized electronic medical record (EMR) system for dentists. By leveraging “voice input + AI analysis,” the system enables automated EMR documentation, effectively helping physicians improve diagnostic and treatment efficiency.

This AI solution meets two key criteria: First, there is a broad market demand, as virtually all clinical departments require voice-based medical record entry. Second, its primary limitation—the adverse impact of noisy recording environments on audio quality—is surmountable. iFlytek has employed a series of measures to mitigate this limitation, such as developing directional microphones and noise reducers to optimize sound capture. This represents a positive trajectory in AI development and explains why iFlytek’s voice AI has been rapidly deployed in over 200 hospitals.

CT-FFR is another example demonstrating a genuine need for AI. CT-FFR provides clinicians with a novel, non-invasive functional assessment tool, enabling evaluations that physicians cannot perform based on imaging alone. Its primary purpose is to enhance physician efficiency, offering innovative clinical value. Currently, Keya Medical, Shukun Technology, and Bodong Medical have all rapidly deployed solutions in this sector, with Keya Medical’s AI product being the first to receive Class III medical device approval from the National Medical Products Administration (NMPA).

After meeting the demand, enterprises need to address commercialization issues. At this point, we must discuss the costs hospitals incur to obtain these services, and answer the question “Why invest in AI?” from the perspectives of both tertiary Grade-A hospitals and primary healthcare institutions.

Taking AI-based imaging products as an example, hospitals need to consider the tangible value they can deliver. If an AI solution merely accelerates physicians’ workflow without generating actual benefits for the hospital, the likelihood of procurement remains low.

Our research indicates that private healthcare institutions are more inclined than public ones to procure advanced equipment, as such devices can effectively reduce their operational costs and enhance patients’ perception of the hospital. Similarly, for AI products to be successfully implemented in hospital settings, it is essential to identify specific stages within clinical workflows where AI can drive cost reduction and improve efficiency.

Outside of hospital settings, the demand for medical AI may not be measured solely by return on investment.

In the post-pandemic era, the national initiative to promote tiered diagnosis and treatment and build a new-generation public health system is accelerating. As part of the “new infrastructure” drive, the state is continuously increasing investment in medical AI. The sustainable development of AI will stem from empowering primary healthcare settings, thereby establishing a medical system with optimized allocation of health resources.

Liu Qingfeng, Chairman of iFlytek and a Deputy to the National People’s Congress, put forward eight proposals during the Two Sessions, two of which focused on empowering primary healthcare and called for top-down reforms.

Specifically, he proposes that the National Health Commission take the lead in leveraging technological tools such as artificial intelligence (AI) to bolster infectious disease prevention and control, starting from the primary healthcare level. Furthermore, he recommends that the Ministry of Science and Technology and the National Health Commission increase research support for medical AI, promote the adoption of AI-assisted diagnostic and therapeutic technologies already deployed at scale—such as “Intelligent Medical Assistant”—accelerate the evaluation and implementation of AI tools like intelligent voice-call robots to support family doctors, and expedite the development of standardization and certification systems for medical AI products to foster a standardized understanding of AI applications.

Liu Qingfeng believes that, “In the face of structural issues such as the relative scarcity of high-quality medical resources and insufficient primary healthcare service capacity in China, AI technology has already achieved significant results in people’s livelihood areas, including improving primary healthcare standards and strengthening healthcare security.” In other words, accelerating the application of AI in China’s primary healthcare sector can not only enhance primary healthcare service capabilities but also play a role in “prevention, control, and treatment before disease onset,” which represents the future value that AI can bring to primary healthcare.

Behind this, AI is needed to help primary healthcare institutions improve medical quality, retain patients, and train young doctors. However, as these institutions often lack financial resources, such an upgrade process should be led by the National Health Commission, as suggested in some proposals at the Two Sessions.

The Path to Success Is Winding and Upward—So Is AI.

Since 1956, AI has experienced three summers and two winters. During this process, scientists have gradually overcome limitations in computing power and various types of data, only to become entangled in new constraints once again, before achieving further breakthroughs. This cycle of advancement and setback has repeated itself, leading to a spiral upward trajectory.

As medical AI continues to advance, it must remain true to its original mission and return to the fundamentals of healthcare. By focusing on needs assessment, practical implementation, and cost analysis, the industry can avoid developing AI products that exist in name only but lack meaningful value.

Revisiting the question, “What medical problems can AI solve?” it may be more appropriate to reframe it within the healthcare sector as “What problems does healthcare need AI to solve?” Only by adopting the right perspective can we unlock broad and valuable growth opportunities.