Ecosystem Co-Creation and Product Integration: A New Path Forward for Medical AI — YiZhan Technology Redefines AI in Medical Imaging

“The most fundamental distinction between humans and animals lies in the human capacity to manufacture and use tools.” From primitive times to the present, we have continuously improved our production tools through the accumulation of experience derived from practice.

Today, the advent of the artificial intelligence era has accelerated the upgrading of productivity tools. We can rapidly aggregate the clinical expertise of renowned physicians to create a digital “expert physician” system, thereby assisting young doctors in their work. In the future, junior physicians will be able to leverage the capabilities of senior physicians, while senior physicians will be freed from routine tasks to pursue more profound medical breakthroughs.

However, the road ahead remains long and arduous. Realizing this vision cannot be achieved solely through advancements in artificial intelligence technology. On one hand, we must continue to effectively collect and leverage vast amounts of medical data; on the other hand, only by implementing structural upgrades within the healthcare sector can these technologies be deployed to primary care settings, thereby enhancing productivity at the grassroots level.

Data shows that China conducts approximately 2 billion radiological examinations annually, with 50% of these being digital radiography (DR) examinations. Physicians spend a significant amount of time each day drafting textual reports to describe patients' clinical findings observed during the examinations and to provide diagnostic opinions.

The interpretation and understanding of medical images are typically performed by professional healthcare practitioners. In a populous country like China, physicians are required to review hundreds of images daily. This workload not only consumes a significant amount of their time, but the process of drafting medical imaging reports is also excessively repetitive and tedious.

The more serious issue is that the majority of patients are located in primary care settings, while high-quality medical resources are concentrated in first- and second-tier cities. This disparity results in an excessive workload being shouldered by insufficient medical service capacity. Physicians remaining in primary care are either newcomers just entering the profession or retired doctors rehired post-retirement. Young physicians, constrained by their professional expertise, struggle to produce efficient and accurate medical imaging reports, whereas older physicians, overwhelmed by hundreds of health examination reports, are prone to overlooking issues within them. Statistical data indicate that the comprehensive misdiagnosis rate in primary healthcare reaches as high as 60%. Such a high misdiagnosis rate significantly undermines the reputation of medical institutions, further intensifying patients’ willingness to seek medical treatment in Beijing, Shanghai, and Guangzhou.

So, can artificial intelligence change this situation?

At the CCR conference, Xi'an Wingspan Electronic Technology Co., Ltd. launched its AI-powered product, W-insight. This solution automatically generates classification and imaging findings reports for 17 thoracic and pulmonary conditions based on digital radiography (DR) chest images, enabling an end-to-end diagnostic workflow encompassing image interpretation, diagnosis, and report generation. It can be widely applied to patient health screening programs, significantly reducing operational costs for health examination institutions and effectively enhancing the quality of primary healthcare services.

Director Liu Shiyuan Attends Wingspan Technology Product Launch Event

At the press conference, Bian Haifeng, CTO of Wingspan Technology, stated, “Compared to existing DR products, most medical imaging AI solutions on the market have focused on breakthroughs in computer vision. However, current product scenarios require concurrent advancements in both computer vision and natural language processing. Each of these fields presents substantial technical barriers, let alone the challenge of integrating them.” Regarding the challenges encountered during product development, he identified three primary sources.

First, most medical imaging AI products on the market adopt a single-task learning framework. Taking pulmonary nodules in chest CT images as an example, the conventional approach is to apply transfer learning by directly adapting well-validated and effective learning frameworks from related fields to the medical domain. However, in the context of digital radiography (DR) products, AI is required to perform multi-label classification for 17 types of thoracic diseases based on images, as well as generate text in the form of a lengthy descriptive paragraph. There are no off-the-shelf algorithms available for this complex task; therefore, researchers must develop a novel deep learning framework capable of integrating multiple tasks.

Secondly, it is highly challenging to localize abnormal image regions and provide accurate descriptions thereof. To address this, Wingspan Technology introduces a co-attention mechanism that simultaneously focuses on the images and predicted disease labels, thereby establishing a synergistic relationship between visual and semantic information.

Finally, descriptions in imaging reports are typically lengthy and comprise multiple sentences, posing significant challenges for existing single-layer LSTM models in generating such long sequences. To address this issue, Xi'an Wingspan Electronic Technology Co., Ltd. has adopted a hierarchical LSTM model integrated with a co-attention mechanism, dividing the generation process into two steps: first generating high-level topics, and then producing fine-grained descriptions based on these topics.

Of course, the advancement of algorithms relies on validation with high-quality data. Over a two-year R&D period, Xi'an Wingspan Electronic Technology Co., Ltd. collected data from nearly 200 hospitals, curating approximately 100,000 high-quality images along with their corresponding reports, thereby gradually overcoming the aforementioned challenges. To date, the company has validated its products using more than 100,000 cases and will continue to refine them through real-world data, striving for near-perfect performance.

In reality, the hardware infrastructure of primary healthcare is not lagging behind; many county-level hospitals have procured advanced equipment such as multi-slice CT scanners and 1.5T MRI systems. However, due to factors including poor hospital location, scarcity of medical personnel, and lack of patient trust, much of this medical equipment remains idle and underutilized.

Therefore, relying solely on artificial intelligence technology is insufficient to resolve these issues. Given the fragmented landscape of domestic medical institutions and the lack of resource connectivity, it is more critical to reconstruct the ecosystem by leveraging underutilized resources and optimizing the allocation of healthcare resources.

In response, Xi'an Wingspan Electronic Technology Co., Ltd. developed the Wingspan Cloud Imaging Medical Image Diagnosis Platform, which leverages cloud platform technology to enable online medical image diagnosis, thereby breaking through temporal and spatial constraints and extending high-quality medical resources to every corner.

This platform provides cloud storage for medical imaging to hospitals, medical consortia, and independent imaging centers, enabling multi-platform access to medical images anytime and anywhere, with mobile phones also supported for image viewing.

Through the regional medical imaging consortium built by Wingspan Technology, patients can undergo imaging examinations at primary healthcare institutions. The acquired images are transmitted to the Cloud Imaging Diagnostic Platform, where they are interpreted by senior physicians from physician groups or tertiary hospitals. Currently, this diagnostic platform has achieved comprehensive coverage across regions, timeframes, and services, providing primary hospitals with uninterrupted 24/7, 365-day-a-year support. It issues diagnostic reports for DR, CT, and MRI scans, with ultrasound and endoscopic imaging capabilities soon to be added. Real-time diagnostic reports are available for night clinics and emergency cases.

Today, Wingspan Technology has deployed its cloud platform across China, with nearly 2,000 hospitals connected to this diagnostic platform. The Wingspan Cloud Imaging Smart Diagnostic Platform enables comprehensive coverage of primary care facilities, reducing missed and incorrect diagnoses while helping radiologists write reports more accurately and efficiently, minimizing repetitive tasks, and significantly boosting productivity in radiology departments.

Furthermore, leveraging the extensive hospital user base on its cloud platform, Xi’an Wingspan Electronic Technology Co., Ltd.’s newly released DR product can be rapidly deployed across these 2,000 hospitals, a feat that traditional AI companies struggle to match.

Wingspan Technology, which started its business in software services in 2009, focuses on the medical field. Its products cover medical imaging, healthcare informatization, and image post-processing, among other areas. To date, it has obtained 154 patents.

Ni Meng, CEO of Wingspan Technology, told VCBeat, “Wingspan Technology has always focused on the B2B sector, with the goal of providing a comprehensive solution to medical challenges. This long-term focus has enabled Wingspan to establish many unique advantages. First, our R&D personnel possess both medical and engineering mindsets, and have integrated these two modes of thinking through extensive practical experience; this constitutes our talent advantage. Second, we can rapidly collect high-quality, exemplary cases through our cloud-based diagnostic platform and submit them to physician groups and medical institutions for annotation, thereby jointly developing AI products; this is our channel advantage. Third, we operate our own open AI platform, which not only supports our proprietary products but also facilitates collaborative product development with other companies. These products can be rapidly deployed to thousands of users via the cloud platform, accelerating commercialization.”

“The significance of our open platform lies in bringing together more partners and institutions to jointly build and develop on our platform. Given the vast array of disease types, no single company can monopolize the entire AI healthcare sector; therefore, we need collaboration to expand AI technology into broader fields,” said Ni Meng, CEO of Wingspan Technology, at the press conference. “Furthermore, we aim to comprehensively extend high-quality medical technologies to primary care settings, ensuring that the general public can not only afford healthcare but also receive effective treatment. This aligns with the national call for a ‘tiered diagnosis and treatment’ system.”

In the future, Wingspan Medical Group will continue to deepen its R&D efforts, leveraging AI technology to empower primary healthcare. By bringing top-tier medical resources to the grassroots level through the Cloud Shadow Smart Imaging Diagnosis Platform, the group aims to better align with the real needs of the public and help patients achieve early detection and early treatment of their conditions.