Five Emerging Trends in Medical AI and Digital Health Following ChatGPT's Breakthrough

“With the rapid advancement of AI technology and the continuous expansion of model parameters, quantitative changes have led to qualitative leaps, giving rise to emergent intelligence,” remarked Liu Junwei, General Manager of Baidu Smart Healthcare, at the Future Medical Technology Conference.

Although 2022 appeared to be a year of cooling sentiment in the digital health sector, with the early-pandemic fervor subsiding from primary markets to IPOs, investors have adopted a more cautious stance.In 2023, the digital technology sector once again witnessed a surge in fervor.Most notably, the industry leader has emerged. OpenAI’s ChatGPT fired the first shot in the deep application of AI.

The underlying reason isIn-Depth Exploration of Technological Value。

Judging by the attendees at the 2023 Future Healthcare Technology Conference, we could also sense the audience’s immense enthusiasm for digital technology. The conference rooms were packed with attendees, including scientists, investors, and corporate business development professionals.

In the words of Wang Guangyu, a professor at Beijing University of Posts and Telecommunications,The development of foundational (large) models has endowed machines with the capabilities to perceive, think, explore, and create, which will have a profound impact on society and industries, thereby attracting significant attention and R&D investment from both the research community and the industrial sector.。

At the conference, Professors Zhang Yuanting, Zhou Shaohua, Wang Guangyu, and Gong Yan, all attending academics, concurred that artificial intelligence will profoundly impact the industry as digital technologies advance. They are respectively conducting frontier explorations in intelligent diagnostic devices, medical imaging, healthcare services, and microscopic imaging.

Meanwhile, in the capital markets, digital technology has undergone a rapid surge in interest. The fervor surrounding ChatGPT has spread from the product itself to various sectors, igniting global enthusiasm starting from Silicon Valley. Google Trends data shows a sharp spike in ChatGPT’s global search interest; from the perspective of the venture capital and private equity community, startups in this sector have continued to attract strong investor demand since the beginning of 2023.

Undoubtedly, this wave of hot air has brought opportunities to all enterprises, while also serving as a breakthrough point for digital technology—Identify genuine application scenarios and aim for rapid implementation.。

Digital technology has become deeply embedded within the healthcare industry. Taking medical services as an example, Elsevier’s “White Paper on the Future Physician” indicates that over the next decade, three major trends will shape global healthcare development: the deep integration of digital technologies with medical services, a comprehensive improvement in patients’ health literacy, and the diversification of healthcare scenarios. Big data will be extensively integrated into population health management. The continuous accumulation of data from scientific research, electronic medical records, and interconnected medical devices will help physicians formulate more precise diagnosis and treatment plans, thereby enhancing decision-making efficiency.

At the Future Technology Conference, we witnessed not only the integration of digital technologies with healthcare services but also their diversified applications in intelligent diagnostics, drug development, and medical imaging.

The integration of digital technology with medical processes offers multifaceted advantages.

From the patient’s perspective, digital healthcare not only transcends time and space to address information asymmetry between patients and providers, but also simplifies medical processes, reduces costs, and improves the overall care experience. From the physician’s perspective, digital healthcare enables the digitization of patient medical records and health files, thereby enhancing the efficiency of disease diagnosis and patient management, and further unleashing medical productivity. From the healthcare institution’s perspective, digital healthcare facilitates refined internal management, leading to improved managerial and service standards.

At the Future Medical Technology Conference, Professor Wang Guangyu shared“Multimodal Foundation Large Models for Healthcare”, the BUPT team, leveraging tens of millions of biomedical data points, conducted research on understanding and generation based on large language models. By integrating medical knowledge with a reasoning core and multimodal intelligent semantic computation methods, they advanced “human-centric” semantic alignment and natural interaction.

Currently, the BUPT team has developed ClinicalBERT 1.2B for general medical scenarios, the large language model ClinicalGPT 175B (with ClinicalGPT 7B-Base released on Hugging Face), and UniBind, a protein function analysis framework based on large-scale pre-trained models (published inNature Medicine). Furthermore, the team has explored TCM-GPT, a multimodal large language model for Traditional Chinese Medicine (TCM), built upon ClinicalGPT. It demonstrates superior performance across multiple dimensions, including TCM theory, syndrome differentiation and treatment, meridian and acupoint knowledge, and herbal compatibility.

Xu Liqun, Chief Scientist at China Mobile Research Institute, also discussed “Opportunities and Challenges in the Transformation of Healthcare Services in the Era of Large Language Models,” whereinAutomated Medical Record Generation SolutionIntegrate large language models into clinical workflows to assist physicians with documentation across various care settings, including outpatient, emergency, and inpatient services.

In addition, Liu Junwei, General Manager of Baidu Smart Healthcare, also shared at the conference Baidu’s applications based on the Lingyi Large Language Model in “Patient-Physician-Pharmaceutical “Regarding practical implementation in these areas, he stated, “Large language models offer higher accuracy and shorter development cycles. They will deliver new outcomes in professional empowerment, quality and efficiency improvement, and enhanced user experience across applications such as intelligent health managers, AI-powered physician assistants, and smart enterprise services, thereby truly bringing new productive forces to the big health industry.”

“Health initiatives advocate for early screening, early diagnosis, and early recovery. While policies are in place, there is still a significant shortage of medical devices,” said Zhang Yuanting, Academician of the International Academy of Medical and Biological Engineering and Founder of the Hong Kong Institute of Medical Engineering.

Therefore, wearable monitoring devices hold significant importance for both consumer (C-end) and business (B-end) users. For consumer users, wearable medical devices will provide usersProvide real-time health monitoring data, enabling users to understand their health status and facilitating scientific health management. For B-end users, the timeliness of wearable medical health devices provides robust medical support for resource allocation in healthcare institutions, allowing physicians to conduct remote consultations and reduce treatment costs.

“Previous wearable monitoring devices were not very convenient,” said Professor Zhang Yuanting. In 2001, Professor Zhang was the first to propose non-wearable intelligent monitoring devices. He also stated thatFuture intelligent wearable detection devices will develop toward miniaturization, intelligence, standardization, and non-intrusiveness.。

The domestic high-end microscope market is currently monopolized by the four major German and Japanese manufacturers. Most domestically produced equipment is characterized by high volume but low quality, with an extremely low market share. Whether from a market perspective or a technological integration perspective,The integration of computational optics and microscopy imaging is an inevitable trend in the development of the market/information age.。

Traditional optical imaging is built upon the principles of geometric optics, adopting the “what you see is what you get” paradigm of human vision, while neglecting many high-dimensional optical information. In the field of microscopic imaging, it fails to simultaneously meet the demands for a wide field of view and high resolution.

In contrast,Computational Optical ImagingGuided by specific application tasks, we acquire or encode light field information across multiple dimensions (such as angle, polarization, and phase) to establish a new sensing paradigm for sensors that far surpasses human vision. Meanwhile, by integrating mathematics and signal processing expertise, we deeply mine light field data to break through the limits of traditional optical imaging.

Gong Yan, a distinguished researcher at the Chinese Academy of Sciences, stated thatOptical microscopy, characterized by its non-invasiveness and broad applicability, is the preferred observational tool in current research fields such as biology, medicine, and pharmacy.In this field, Professor Gong Yan’s team has comprehensively upgraded the full-chain digital manufacturing processes encompassing design, machining, assembly integration, and system testing, and has independently developed multiple series of high-end microscope objectives.

These include the development of a STED–two-photon hybrid microscope prototype (achieving an ultra-high resolution of 47 nm and an imaging depth of 141.5 μm); focusing on asymmetric three-beam interference illumination combined with segmented half-wave plates to enhance imaging speed; and implementing large-field-of-view optical tomography microscopy to increase the spatial resolution of the imaging system to hundreds of megapixels.

In the future, high-end microscopes will continue to pursue higher resolution, faster imaging speeds, and greater imaging depth. Professor Gong Yan noted that emerging technologies will be integrated with optoelectronic microscopy in the years ahead.

One isNovel Optical Devices and Light Field Modulation Mechanisms. Novel devices such as metalenses, metamaterials, plasmonics, and photonic crystals provide new means of optical control for computational optics. Another isEmerging Mathematical Algorithms and Computational Performance. To develop new theories, mechanisms, and technologies for various complex real-world imaging applications, establish optimized multi-parameter design and optical regulation mechanisms, and provide strong support for the advancement of computational imaging.

Another prominent application of digital technology is in medical imaging. The global surge in popularity of software such as ChatGPT has demonstrated the capability of generative AI to produce text or images from complex user prompts. With broad application prospects, this technology has already begun to achieve significant results in the healthcare sector.

In 2023, Siemens Healthineers showcased innovative concepts and technical prototypes based on generative AI. Unlike other companies that focus on text or image generation, Siemens Healthineers enables users to quickly locate and highlight corresponding areas in reports by clicking on medical images, through the loading, linking, and preparation capabilities of its intelligent chat system. More notably, Siemens HealthineersLeverage AI to dynamically generate diagnostic imaging reports and prioritize them by clinical significance, enabling physicians to process information more efficiently.。

At this conference, Professor Shaohua Zhou, Chair Professor at the University of Science and Technology of China and Member of the National Academy of Inventors, outlined two potential forms of AI-generated medical imaging: one is medical image restoration, and the other isMedical Image Synthesis。

Currently, Professor Zhou Shaohua’s team is exploring new methods for synthesizing MRI images. “We measure various parameters, including demographics, genomic sequencing data, biomarkers, and imaging data. It is evident that causal relationships exist among these categories of measurements. The resulting images reflect both demographic information and biomarker data.” By leveraging causal models to regulate different parameters, this approach also transcends spatiotemporal limitations.Images from a decade ago can also be synthesized using computer-based methods.。”

Drug discovery and development are key research areas for pharmaceutical companies and medicinal chemists. However, inefficiency and high costs have long been obstacles in this field.

It is widely acknowledged by many professionals engaged in drug research and development that the digital era is transforming the pharmaceutical industry. Machine learning and deep learning algorithms have been applied to various stages of the drug discovery process, including peptide synthesis, virtual screening, toxicity prediction, drug monitoring and release, pharmacophore modeling, quantitative structure-activity relationship (QSAR) analysis, drug repurposing, and the assessment of multiple pharmacological and physiological activities.Significantly Shorten Drug Time-to-Market and Enhance Safety。

Wang Taifeng, Head of AI Algorithms at BioMap, is another example. At the Future Medical Technology Conference, Wang shared insights on “AI Foundation Models for Target Discovery and Drug Design.” He stated that natural language alone is insufficient to help pharmaceutical companies leverage more information. Therefore, BioMap has developed aBillion-Scale Protein Language Large Model, leveraging this model to advance protein prediction and design, and further utilizing AI-driven drug discovery systems to screen compounds for new drug development.

Above, we have described the integration of digital technology with multiple application scenarios. Despite its diverse innovations, immense power to create miracles, and boundless creativity, it has failed to escape a critical bottleneck:How Digital Technologies Can Be Rapidly Implemented While Ensuring Information Accuracy and Security？

In terms of technological commercialization, China’s large language model (LLM) industry is currently plagued by intense involution. In contrast, foreign vendors place greater emphasis on the practical deployment and application of LLMs.From intense competition in the technology itself to fierce rivalry in application scenarios, this is an inevitable stage in the commercialization of China’s data technology.。

Regarding information accuracy, Wang Guangyu stated: “General-purpose large language models still severely lack domain-specific knowledge.“Although ChatGPT can, to some extent, answer medical questions, it remains significantly deficient in highly specialized areas, such as optimizing the diagnosis of patient symptoms or formulating treatment plans. Current large language models still lack genuine professional reasoning capabilities, as well as the domain-specific accuracy, compliance, and safety required in healthcare.” There remain numerous technical challenges to be explored and resolved in the transition from general-purpose foundation models to specialized large models for the health sector.

And on the issue of moral ethics,The training and application of generative AI require large amounts of patient data, which may raise concerns regarding data privacy and confidentiality.. To ensure the security of patient information, healthcare institutions and enterprises must implement strict data management measures and comply with relevant laws and regulations.

Revolutionary Change Often Coexists with Risk. Although digital technology is facing certain challenges and ethical concerns, industry professionals are addressing the field with a focus on safety, fairness, evidence-based practice, and privacy. With proper planning and management, digital technology is poised to propel the healthcare industry into a new era characterized by greater efficiency, precision, and personalization.