From Routine to Precision Pathology: Intelligence and Cloud Integration as the Future Trends of the Pathology Industry [2019 VB100]

On December 20, 2019, the “2019 Top 100 Future Healthcare” Forum, hosted by VCBeat and Eggshell Research Institute, and co-hosted by KPMG China, Legend Capital, BV Baidu Ventures, Weilai Capital, Puhua Capital, Tsinghua Nomura China Research Center, Changling Capital, Legend Star, Yuanjing Capital, the Internet Hospital Branch of the Chinese Society of Research Hospitals, Aimeda, Zero2IPO Capital, and Yanzhi, grandly opened at the Jiuhua Villa in Beijing.

As a dedicated observer and chronicler of the innovative healthcare industry, VCBeat officially launched the Future Healthcare 100 Forum in 2015. Now in its fifth year, the event has gradually grown into a premier brand activity within the innovative healthcare sector.

At the Digital Pathology and Precision Diagnosis Forum, Professor Bu Hong from the Department of Pathology at West China Hospital of Sichuan University; Zhou Mi, Managing Partner at Puhua Capital; Qi Hua, CEO of Sinotest Biology; Gao Zhibo, CEO of Yuce Bio; Zhang Yang, Co-founder, Executive Director, and President of Hengdao Pathology; and Huang Chunbo, General Manager of Huayin Health, delivered insightful speeches and jointly explored the development prospects of the digital pathology and precision diagnosis industry.

Prof. Hong Bu, West China Hospital

Intelligence and cloud-based transformation are inevitable trends in the future of pathology. Intelligence signifies the broader development of artificial intelligence, while cloud-based transformation represents the shift of pathology from individual labor to clustered, collaborative work. Pathological diagnosis can be standardized, reproducible, quantifiable, and normalized, rather than relying solely on morphological assessment and empirical understanding. However, a significant challenge currently facing digital pathology is suboptimal “user experience,” which often overlooks the actual workflow scenarios of implementation. Without a convenient workflow and a positive user experience for pathologists, it will be difficult for intelligent and cloud-based solutions to be effectively deployed in clinical practice.

Digital pathology will undergo three stages of development:

1. Telepathology—the initial stage of digital pathology—is currently undergoing rapid development.

2. The digitization of certain pathology department data has enabled the integration of artificial intelligence and intelligent pathology into the field, which is currently in an exploratory phase within the industry.

3. Industry promotion and capital intervention must work in tandem to drive the development of intelligent pathology and cloud-based solutions.

Intelligentization and cloud-based transformation are inevitable trends in the development of pathology departments, at least because:

1. The current state of pathology is characterized by a shortage of personnel, a situation that is expected to persist for a considerable period. There is a particularly acute scarcity of pathology technicians, which has not received sufficient attention from the industry. Future development prospects, as indicated by educational backgrounds and professional titles, are also quite pessimistic, with the vast majority of pathology staff holding junior or intermediate-level positions. With the rapid growth of immunohistochemistry and molecular diagnostics, the demands on pathology personnel will increase, along with their workload. The shortage of pathology staff will become even more severe in the future, making the adoption of intelligent technologies to improve work efficiency an inevitable path forward.

2. The high difficulty of pathology subspecialization, due to a shortage of pathology personnel, with the average number of pathologists in most pathology departments being only 2–3, makes it challenging to promote subspecialization, resulting in pathological quality failing to keep pace with clinical advancements. Issues related to subspecialization must be addressed through future collaboration and resource sharing, namely, by adopting a cloud-based model.

3. The operational framework urgently requires optimization. We typically categorize the operational framework into three broad tiers: national-level teams and regional centers, other tertiary hospitals, and secondary hospitals. In reality, the landscape of pathology remains dominated by the pathology departments of Grade 3A hospitals, and we have yet to identify effective strategies for advancing pathology services at the primary care level. Compared with international standards, the number of third-party laboratories is not excessively large; rather, their development is insufficient and their quality remains suboptimal. Third-party laboratories operate in a cross-regional, multi-site model, which necessitates greater cloud-based integration.

Scene.

These pathological challenges at three levels all require the application of intelligent pathology and cloud-based technologies. Future pathology should not be limited to interpreting self-generated data, but rather focus on interpreting and integrating data across the entire disease spectrum. While the technical implementation in real-world pathological scenarios may begin with mathematical approaches—an area where engineers excel—the ultimate output must be grounded in medicine, translating into clinical actions through pathologists. This represents a collaborative future.

Zhou Mi, Puhua Capital

From a historical perspective, in vitro diagnostics has evolved from experimental biology toward molecular biology. Initially, diagnostic approaches focused on morphological, histological, or biochemical levels. In recent years, increasing attention has been directed toward immunology, cytology, and molecular biology, aligning with the trend toward precision medicine. Against this backdrop and supported by emerging technologies, pathology is transitioning from routine pathological examination to specialized testing. Routine examination primarily refers to hematoxylin and eosin (H&E) staining, while immunohistochemistry and in situ hybridization are rapidly gaining prominence within specialized testing. In other words, pathological methods are shifting from early cellular staining techniques toward approaches based more heavily on molecular pathology and disease subtyping. Consequently, pathological diagnosis is progressively advancing from a generalized state to one that is personalized and precise.

With the rapid development of artificial intelligence, big data, and cloud computing, medical AI gradually became a hot investment area around 2015. Initially, many medical AI companies entered the market through medical imaging; however, artificial intelligence is now driving the rise of automated pathology-assisted diagnosis. More than 50% of medical AI companies are developing pathology-related applications centered on diagnostic indicators, genotyping, and disease history.

From an investment perspective, the pathology industry’s value chain comprises three segments:

1. Manufacturers of Diagnostic Instruments and Reagents

2. Upstream to Third-Party Pathology Diagnostic Centers for Oncology

3. Public and private medical institutions of all types

Within the value chain of the entire pathology industry, effective data accumulation is key to the development of AI in pathology—a consensus widely shared across the sector. The overall process of developing AI applications for pathology can be divided into two parts: accumulating data from hardware vendors, hospitals, and third-party testing laboratories, and algorithm development. Algorithm development itself also consists of two components: one involving groups of pathology experts, including medical associations and physician societies; the other involving enterprises specializing in the development of pathology algorithms. The final implementation scenario will integrate pathology expertise with algorithm development before returning to hospital pathology departments and third-party testing institutions.

The development of pathology AI is still in its early stage of data accumulation. As an auxiliary tool, pathology AI is integrated with physicians’ diagnostic workflows. This implementation approach carries low risk and has become a new global trend in pathology. Digital pathology companies have also attracted attention from the most traditional players in the pharmaceutical industry, particularly developers of targeted therapies. For example, in November 2019, digital pathology company PathAI completed a $75 million Series B financing round, with additional investments from Bristol Myers Squibb (BMS) and Merck & Co. (MSD).

Overall, the development of the pathology industry is primarily driven by three factors: the accelerated R&D of oncology drugs has unleashed demand for diagnostic services within the pathology sector; policy support has facilitated the clinical implementation of AI in healthcare; and the severe imbalance between the supply and demand of medical resources has spurred the development of AI-assisted pathological diagnostic tools. The future of the pathology industry will be characterized by two key trends: chain-based and industrialized development, and industry upgrading empowered by AI.

Sainote Bio Qi Hua

From the perspective of clinical practice, pathological diagnosis is the “gold standard” for disease diagnosis. From the hospital’s standpoint, the capabilities of its pathology department reflect the institution’s overall diagnostic and therapeutic proficiency. Due to the critical importance of pathological diagnosis and the uneven distribution of pathology resources, pathology departments and pathological diagnostics face significant challenges as well as opportunities. As malignant tumors have become a major public health issue threatening human health, the persistent high incidence of cancer presents substantial opportunities for the pathology industry.

The pathology industry has evolved from histopathology to cytopathology, immunopathology, molecular pathology, and digital pathology. Liquid-based cytology, as a representative of cytopathology, has been widely adopted in primary care hospitals. However, cytopathology extends far beyond liquid-based cytology; specimens such as those obtained from fine-needle aspiration, cerebrospinal fluid, and pleural or ascitic fluid represent important application scenarios for cytopathology.

Over the past two decades, immunohistochemistry (IHC) has experienced rapid growth in the field of pathology. In a sense, IHC serves a role in pathology departments analogous to that of chemiluminescence assays in clinical laboratories. IHC plays a critically important role in pathological differential diagnosis, definitive diagnosis, prognosis assessment, and companion diagnostics, making it a highly favored sector among investors in the pathology industry.

Currently, immunohistochemistry is primarily a qualitative and semi-quantitative assay rather than an absolute quantitative one; this limitation is attributable to the use of enzyme-catalyzed substrates. In the future, immunohistochemistry will evolve from a qualitative and semi-quantitative method into a tool for absolute quantification.

Currently, molecular pathology is the most sought-after field in pathology, represented by techniques such as FISH, PCR, and high-throughput sequencing. However, pathological diagnosis is comprehensive, integrating morphology, immunology, and molecular biology; therefore, pathological conclusions are not based solely on molecular diagnostics.

Molecular diagnostics has a significant application scenario in liquid biopsy. Due to the challenges associated with tissue sampling for histopathology, liquid biopsy tests, represented by ctDNA and other markers, hold substantial promise for future applications in pathology departments.

Currently, cancer treatment has entered the era of personalized therapy, which is accompanied by precision diagnosis. In clinical practice, methods such as immunohistochemistry have become widely used to achieve precise diagnosis for targeted cancer therapy. One concrete manifestation of this approach is formulating cancer treatment strategies by detecting gene or protein targets. Furthermore, with the in-depth application of various techniques—including tissue staining, immunohistochemistry, PCR, FISH, and NGS—in pathological diagnosis, precision diagnosis will play an even greater role in cancer treatment. In the future, cancer classification will evolve beyond mere histological categorization to include classification based on molecular biomarkers. Pathological diagnosis will also shift its focus from cancer diagnosis to cancer prevention.

Yuce Bio, Gao Zhibo

The most significant breakthrough in the era of precision medicine lies within tumor immunology. Hot areas in this field include PD-1/PD-L1 monoclonal antibodies, T-cell receptors (TCRs), and personalized RNA vaccines, all of which share common mechanisms. The logic of immunotherapy can be broadly categorized into two approaches: immune normalization and immune enhancement. According to the cancer-immunity cycle theory, tumor immunity involves key steps such as the release of neoantigens, activation of antigen-presenting cells and T cells, T-cell infiltration, and the recognition and killing of tumor cells. A disruption at any of these steps can impair the proper functioning of the immune system; therefore, restoring immune normalcy is a crucial strategy, with existing PD-1/PD-L1 inhibitors serving as typical examples. Meanwhile, personalized vaccines and cell therapies represent effective methods for enhancing immune responses.

In the immunotherapy market, antibody drugs targeting PD-1/PD-L1 are experiencing rapid growth in market promotion and widespread clinical application. For patients who fail to respond to PD-1/PD-L1 therapies, personalized vaccines and cell therapies represent important treatment modalities, and this direction is poised to become the next major growth hotspot in the future.

An analysis of clinical trial distributions reveals that combining pembrolizumab (Keytruda) and nivolumab (Opdivo) with other antineoplastic agents to enhance immunotherapeutic efficacy has become a prevailing trend. However, inadequate rigor in clinical trial design may concurrently lead to failures in demonstrating the clinical efficacy of these combination regimens.

There are two potential pathways for the clinical implementation of personalized vaccines and cell therapies: one is to regulate them as drugs, and the other is to classify them as Category III new technologies in hospitals. Overall, the landscape of immuno-oncology diagnostics and therapeutics in China can be categorized into three segments: numerous companies focusing on PD-1/PD-L1 targets, a subset of enterprises researching cell therapies, and a significant portion dedicated to the development of personalized vaccines. These rapidly advancing technologies also face various challenges. The most significant challenge in clinical practice is insufficient clinical benefit and low efficacy. There are two approaches to addressing low efficacy: first, identifying responsive patient populations through precise diagnosis; second, developing precise combination treatment regimens. Furthermore, there are additional detailed issues, such as severe side effects, drug resistance, and hyperprogression.

The primary challenge in the research and development of personalized cell therapies and vaccines is inaccurate target identification. Personalized vaccines face three key technical bottlenecks: target screening, cell preparation, and delivery routes, with the latter needing to account for the tumor’s immunosuppressive microenvironment. Tumor neoantigens represent a key direction for personalized vaccine development; however, neoantigen detection faces numerous challenges. The most significant challenge is the high cost and lengthy timeline associated with large-scale target validation, which stems fundamentally from limited validation data, data insufficiency, and the lack of standardized validation protocols. Inaccurate target identification in such assays directly compromises downstream quality and efficacy.

Zhang Yang, Hengdao Pathology

The pace of artificial intelligence (AI) development, particularly after 2010, has been exceptionally rapid. In 2006, Geoffrey Hinton, known as the “father of deep learning,” published a paper in Science that introduced the concept of deep learning, after which deep learning frequently emerged across the technological landscape. Prior to 2011, statistical learning dominated AI algorithms; however, its limitations and shortcomings gradually became apparent in practical applications. In contrast, deep learning demonstrated clear advantages in tasks such as recognition and discrimination. From 2012 to 2015, the overall iteration speed of deep learning accelerated, and its accuracy improved steadily, ultimately surpassing human-level performance by 2015. In 2016–2017, AlphaGo sequentially defeated world Go champions Lee Sedol and Ke Jie. With the three pioneers of deep learning—Geoffrey Hinton, Yann LeCun, and Yoshua Bengio—receiving the 2019 Turing Award, the application of deep learning in artificial intelligence gained comprehensive recognition.

Currently, deep learning approaches have been adopted across various fields, ranging from robotics and security to autonomous driving and medical imaging. However, in the field of pathological artificial intelligence (AI), a significant challenge for deep learning lies in addressing the pathological “gold standard.” Specifically, there is a substantial gray area between cancerous and non-cancerous tissues in pathological diagnosis. Relying solely on AI for judgment in this ambiguous zone would lead to a high rate of misdiagnosis, as it requires subjective clinical assessment by physicians. Therefore, throughout its foreseeable development, pathological AI will remain an auxiliary tool for doctors.

From the perspective of international trends in pathology AI, major countries are actively promoting the development of artificial intelligence in pathology and related data. In 2017, the NHS proposed establishing a network of 29 pathology hubs and spokes to provide pathology services across the entire UK healthcare system. In 2019, the UK Department for Business, Energy and Industrial Strategy proposed the establishment of five Digital Pathology and Imaging Centres to accelerate the pace and progress of AI adoption in pathology. Philips, General Electric, Siemens, and Leica will jointly lead these five new AI-driven digital pathology imaging centres.

In 2005, the U.S. National Cancer Institute launched The Cancer Genome Atlas (TCGA) project to consolidate genomic and biospecimen data. Subsequently, the National Institutes of Health (NIH) introduced the Genomic Data Commons (GDC) initiative based on TCGA, integrating all clinical and genomic data onto a single platform. This platform accepts data from cancer genomics projects worldwide and enables broad sharing of these data with qualified researchers.

Domestic companies are also rapidly and comprehensively deploying in pathology AI, particularly in the field of cytology. While international pathology AI initiatives emphasize drug development and exploratory applications in assisted diagnosis, China’s efforts are more focused on addressing the shortage of primary care physicians and bridging gaps in their clinical skills.

From the practice and exploration of Hengdao Pathology, it can be seen that artificial intelligence can play a catalytic role in its integration with third-party pathology diagnostic institutions, mainly manifested in three aspects:

First, at the micro level, third-party services represent the optimal application scenario for medical AI, as they can continuously improve diagnostic efficiency while reasonably controlling costs.

Second, at the meso level, the multi-center, multimodal nature of third-party data can better and more accurately meet the needs of AI development;

Third, at the macro level, it serves as foundational infrastructure for the research, development, and application of medical artificial intelligence, leveraging advantages in third-party data, annotation, and scenarios.