How Diagens-B (02526.HK) Builds a Medical Imaging Model Factory 04.09.2026, 11:54 Uhr von EQS News Jetzt kommentieren: 0

AI for Science (AI4S) is reshaping the paradigm of life science exploration. Spanning drug discovery, genetic analysis and clinical diagnosis and treatment, AI has emerged as a core engine driving breakthroughs in biological research. In medical imaging, a pivotal track for healthcare AI innovation, AI4S has evolved beyond image reading assistance to build generalized underlying intelligence infrastructure for clinical and scientific research.

As a distinctive practitioner in this global trend, Diagens Technology Co., Ltd.  (02526.HK, “Diagens Tech”) does not engage in pharmaceutical R&D, but builds fundamental AI research infrastructure to empower life science imaging analysis. The launch of the world’s first foundational medical imaging model iMedImage® and the end-to-end platform iMedLoop™ enables hospitals and research institutions to share AI production capabilities through projects and co-development.

Most importantly, Diagens Tech has established a commercial closed-loop featuring “data – model development – product – clinical practice – data”, which underpins Diagens Tech’s one-of-a-kind medical imaging AI Model Factory, enabling in-depth scientific research and sustainable commercial resilience.

Data Standardization: Building a Data‑Governance System amid Complex Clinical Conditions

Medical imaging data is inherently heterogeneous, a challenge further amplified by China’s diversified clinical settings. Different levels of healthcare facilities nationwide deploy equipment from diverse domestic and international brands and models. Coupled with China’s vast territory and large population, the country has cultivated unique patient cohorts and disease spectrums that are hard to be replicated elsewhere. Direct model training on unprocessed raw data would substantially amplify noise and disruption.

Standardization serves as the very first step to convert raw imaging data into high-quality data assets. Powered by intelligent annotation and expert quality control mechanisms, Diagens Tech’s iMedStudio™ delivers multi-layered data refinement through AI precise segmentation, intelligent arbitration and manual expert review, converting raw medical images into standardized training samples. As of end-June 2026, this high-precision data processing pipeline has accumulated approximately 28.95 million annotated samples, supported by a professional team of over 3,000 specialized annotators.

Constrained by clinical data security protocols and on-site data collection requirements, standardized medical imaging infrastructure cannot be established overnight. Through nearly a decade of in-depth hospital collaboration, Diagens Tech has fully operationalized its optimized end-to-end data governance pipeline. Its standardized framework eliminates format and annotation inconsistencies while preserving cohort and device diversities. Such heterogeneous data features were deemed constraints in the traditional “one model per disease” approach, yet constitute core advantages for foundational large models. The high-diversity data assets refined from complex real-world clinical scenarios form the cornerstone of robust cross-scenario generalization capabilities.

Model Scaling: Transforming Model‑Building from Craftsmanship to Industrialized Production

This revolutionary shift in production methodology originated from Diagens Tech’s forward-looking strategic decision in 2017. While industry peers prioritized rapid iteration of disease-specific models, Diagens Tech embarked on a long-term, high-barrier path of independent R&D for medical imaging foundational large models. Built on self-accumulated clinical data assets, this proprietary foundational infrastructure cannot be purchased or rapidly replicated, granting Diagens Tech a multi-year technological lead in foundational model development.

Today, the flagship iMedImage® foundational medical imaging model features 104 billion parameters, trained on over 80 million medical images covering 19 mainstream imaging modalities. With this mature foundation in place, new specialty-specific models no longer require full-cycle training from scratch, and can be rapidly deployed via targeted fine-tuning with specialty-specific data. Previously requiring years of data accumulation and iteration, the deployment of specialty-specific models is now compressed to two to three months. As of end-June 2026, Diagens Tech has delivered 158 specialty-specific model projects through cooperation with nearly 100 hospitals nationwide.

The interim‑period results deliver quantifiable proof of operational returns. In the first half of 2026, Diagens Tech’s model service revenue reached RMB 94.541 million, representing a year-on-year increase of 101.1% and accounting for 86.9% of total revenue. R&D expenditure stood at approximately RMB 64.12 million, up 67.4% year-on-year. The substantial outperformance of revenue growth over R&D investment growth validates accelerating platform-based economies of scale. Sustained R&D investment underscores ongoing expansion, with industrialized productivity yet to reach full potential.

The essence of scaling lies in optimized cost structures. The foundation requires only one-time massive investment, supporting iterative development of unlimited specialty-specific models without repeated high-cost input. Multi-project deployment in parallel enables all online models to benefit synchronously from each foundational model iteration. Rather than relying on individual models, value is accumulated across the entire pipeline, fundamentally transforming medical AI model development from craftsmanship to industrial manufacturing.

Replicable Capabilities: Turning Model‑Building into a Reusable On‑Demand Service

Beyond internal production capabilities, Diagens Tech’s Model Factory is evolving toward service-oriented openness, productizing its mature “foundational pre-training + specialty-specific fine-tuning” paradigm as replicable, accessible services for hospitals and academic research institutions worldwide.

This system is underpinned by three core strengths. First, full-process productization. Launched in July 2026, the iMedLoop™ platform solidifies the entire industrial workflow covering data governance, model training, performance evaluation, commercial deployment and clinical feedback iteration. Partner institutions can access complete industrial AI production capabilities via project cooperation and co-development, eliminating the need for in-house pipeline development. Second, cross-modal replicability. Originating from chromosome karyotype analysis scenarios, Diagens Tech’s methodologies have been successfully replicated across 19 imaging modalities, unbound by any specific disease indication. Third, global market accessibility. The full product portfolio complies with NMPA, FDA and CE requirements, with sales networks covering more than 70 countries and regions across six continents.

Over the past decade, Diagens Tech has accomplished a groundbreaking transformation rooted in China’s complex real-world clinical ecosystem: upgrading medical imaging AI development from fragmented craftsmanship to a standardized, scalable and exportable Model Factory. With the September Stock‑Connect eligibility window drawing near, this platform‑driven model factory will come onto the radar of mainstream mainland institutional investors for portfolio allocation. As a global industrial‑grade platform for medical‑imaging AI, Diagens Tech will see its scarce market positioning continuously re‑evaluated by southbound capital once it gains Stock‑Connect eligibility.