Desai Technology 2026 Interim Results Presentation

DeShield-B (02526) 2026 Interim Results Conference | AI Key Takeaways

Key Takeaway: The core message conveyed in this conference is that global medical imaging AI remains in the early stages of rapid development. In the first half of 2026, DeShield’s revenue grew by 21.0% year-over-year, with model services revenue surging by 101.1%, becoming the primary growth driver for the period. The company has preliminarily established an integrated commercialization framework linking its iMedImage foundational medical imaging large model, iMedLoop platform, and products that have obtained medical device registration certificates—forming a cohesive 'foundation–platform–product' ecosystem. Management stated that the next step will focus on deepening collaborations with hospitals domestically and internationally to advance training and clinical deployment of more specialty-specific medical imaging AI models, and to accelerate market adoption of AI AutoVision following its certification.

1. The medical imaging AI market remains in the early phase of rapid expansion

During the conference, management cited forecasts indicating that the global medical imaging AI industry is expected to grow from approximately RMB 10 billion in 2026 to over RMB 150 billion by 2030.

Currently, there are over 3,000 medical imaging diagnostic tests globally, which expand to more than 14,000 when categorized by indication and specific technical pathways; by 2030, AI-powered medical imaging diagnostic tests could exceed 30,000. Management believes that industry opportunities stem not only from applying AI to existing diagnostic tests but also from using AI to create entirely new imaging diagnostic methods.

2. Model services have become the company's primary growth engine

In the first half of 2026, the company’s revenue increased by 21.0% year-over-year, with model services revenue surging by 101.1% year-over-year, making it the main driver of growth during the period. Management stated that the company’s gross margin remains at a relatively high level within the industry.

This indicates that the company’s foundational models and platform capabilities have begun to consistently translate into model services revenue. Meanwhile, as the industry is still in an early investment phase, the company will continue to increase R&D investments in computing power, data, and underlying technologies.

3. The platform’s value primarily lies in the scalable production efficiency of specialty models

The company has preliminarily established an integrated commercialization framework—spanning 'foundation model, platform, and product'—centered around its iMedImage foundational large model, iMedLoop platform, and products that have obtained medical device registration certificates.

According to management, compared with traditional development approaches, building specialty models based on iMedImage reduces data dependency to approximately 1/200 of the original amount, cuts development costs and computing expenses to roughly 1/10, and shortens the development cycle from two to three years down to two to three months.

As of the reporting period end, the company had cumulatively launched 158 model projects and amassed a repository of over 28.95 million high-quality medical imaging cases. This project pipeline and data accumulation provide a solid foundation for continuously developing additional specialty models.

4. Model services are exploring expansion from joint development into recurring per-case revenue streams

The company offers model services through both cloud-based and on-premises delivery models. Management explained that each model project typically involves two phases: Phase one entails co-developing proprietary models with hospitals or enterprises; phase two involves promoting the completed models across hospital consortia or relevant healthcare institutions to generate ongoing revenue through shared intellectual property rights and per-case billing arrangements.

Management illustrated with a representative hospital in Northeast China as an example, noting that the two parties adopted a per-case settlement model. Following the collaboration, the number of patients undergoing relevant tests at the hospital increased year-over-year, and the waiting time for test reports was reduced from over 20 days to approximately 4–6 days. This case demonstrates how the model, once integrated into clinical workflows, can enhance diagnostic and treatment efficiency and establish a per-case revenue stream.

5. AI AutoVision enters the commercial validation phase following regulatory approval

AI AutoVision is primarily used for karyotype analysis of chromosomes and received Class III medical device registration from China’s National Medical Products Administration on May 20, 2026. Management stated that after regulatory approval, the product still needs to go through processes such as parameter listing on procurement platforms, hospital budgeting, bidding, and on-site implementation. The company expects meaningful volume uptake to begin in the second half of this year, with further growth anticipated next year. The predecessor product, AutoVision, has already been deployed in over 400 medical institutions in China, primarily tier-3 hospitals, all through formal bidding processes. The company noted that this existing business gives it deep insight into hospital procurement practices and purchasing cycles.

6. Specialized model architecture and proprietary data form a dual-layer barrier

Management believes that medical imaging requires identifying a large number of abnormalities that follow a long-tail distribution, and current general-purpose image models lack sufficient domain-specific diagnostic capabilities.

The company summarizes its core competitive barriers in two layers: first, a model architecture specifically redesigned and trained from scratch for professional medical imaging interpretation; second, high-quality medical imaging data accumulated over nine years, along with professional annotation and quality control capabilities requiring involvement from highly skilled physicians. iMedLoop and iMedStudio further enhance the efficiency of specialized data generation, annotation quality control, and model training.

Note: The above content was compiled by AI based on statements made during the conference. Market size figures cited reflect management’s on-site references to forecasted estimates. For specific information regarding business plans and future growth expectations, please refer to the company’s official announcements and disclosures.

33K Views