[For more diverse corporate information on the startups mentioned in this article, please visit the Unicorn Factory big data platform 'Data Lab'.]
"The emergence of hallucination issues in medical AI is actually a new opportunity for us. This is because the importance of high-reliability medical data and specialist feedback will inevitably grow even more."
Yoon Seung-gyu, CEO of Mediwork AI, recently made these remarks in an interview with MoneyToday's startup media platform, Unicorn Factory. Launched last October, Mediwork AI is a medical AI data infrastructure company that supports everything from securing and processing the data needed for medical AI development to evaluating and enhancing model performance.
Mediwork AI was established when a subsidiary of the AI data company CrowdWorks, which handled its medical data business, spun off to strengthen its expertise. At the time, Yoon (CEO), who led the medical data business, along with internal director Kim Jae-young, CMO (Chief Medical Officer) and an internist, currently lead Mediwork AI.
Mediwork AI's core competitiveness lies in its network of over 1,200 specialists built even before the spin-off. It has secured a pool of specialists covering 25 medical departments. Based on this, it secures high-quality data required by medical AI companies and performs labeling and processing so that AI can learn. Subsequently, it supports the process where specialists directly evaluate the results of models created by medical AI companies using the data and provide feedback for fine-tuning.
CMO Kim stated, "Even if an excellent AI model is created technically, whether the results are medically accurate and usable in actual clinical settings is an area requiring specialist opinions." He added, "As medical AI becomes more advanced, the role of medical staff with expertise in each field will become even more important."
The starting point for Mediwork AI was the gap between medical AI companies and specialists. Medical AI companies needed to provide medical data for learning and required specialized personnel for processing, but it was difficult to directly recruit suitable medical staff for each project. Conversely, doctors rarely had opportunities to utilize their long-accumulated data and expertise outside of clinical practice.
CMO Kim explained, "From the perspective of doctors, this structure allows them to generate additional revenue by utilizing specialized knowledge that remained confined to clinics, while companies can secure necessary medical expertise." He continued, "Initially, some specialists participated in various labeling and processing projects and earned over 80 million won annually, leading to word-of-mouth among medical staff and enabling the construction of the current broad pool of specialists."
Mediwork AI utilizes this secured network of specialists to obtain high-quality data required by medical AI companies. Specifically, when a medical AI company presents the data it needs, Mediwork AI identifies and connects with medical staff who possess that specific data. It then plans joint research and supports companies in utilizing the data through formal procedures such as Institutional Review Boards (IRB) and Data Review Boards (DRB).
The company takes on the role of coordinating all these processes, from connecting specialists to designing research with hospitals and handling data utilization procedures. Afterward, it processes the secured raw data into refined data that AI can learn through labeling and segmentation tasks, such as marking disease status or distinguishing the location and area of lesions in images.
Currently, the company's core revenue source is also the medical data processing business. The structure involves receiving data projects from companies, connecting suitable specialists to perform the work, and collecting service fees. The core of this business lies in how accurately and quickly it can connect medical staff with the required expertise for each project. As medical AI becomes more advanced, the required expertise is becoming increasingly specialized. Even within the same department, the medical staff available to participate in tasks varies depending on their sub-specialty.
CMO Kim emphasized, "Projects are becoming so specialized that even within gastroenterology, there are tasks that can only be performed by doctors who have majored in specific fields such as the liver and biliary tract." He added, "The company's competitiveness lies in its ability to quickly find medical staff matching requirements and deploy them to projects."
Currently, the area Mediwork AI is focusing on as its next growth driver (core revenue source) is the "medical AI model verification" business. This involves specialists directly evaluating whether the results provided by medical AI companies are medically and clinically valid and assessing if there is any risk when applied to actual patients. Errors discovered during this process are fed back into the AI to fine-tune the models and improve accuracy.
Yoon (CEO) stated, "Especially as hallucination issues in medical AI become more prominent, the process of specialists verifying AI results and providing feedback on incorrect parts will become even more important." He expressed his expectation that "as medical AI becomes more advanced, the demand for such specialist verification will also increase accordingly."
In the medium to long term, the company plans to expand its business to support regulatory approvals from the FDA (U.S. Food and Drug Administration) and the Ministry of Food and Drug Safety for medical AI companies. For medical AI to be approved as a medical device, it must be verified not only on model performance but also on what data was used for learning and whether that data and development process are trustworthy.
In addition, the company plans to utilize its network of over 1,200 specialists to evaluate whether AI model results are medically valid and develop capabilities to create verification materials that can be utilized for obtaining regulatory approvals in the future. Through this, Mediwork AI aims to assist medical AI companies throughout the entire process of obtaining regulatory approval.
Yoon (CEO) said, "Ultimately, the company's goal is to become a 'necessity' in the medical AI industry, such that every company developing medical AI must go through Mediwork AI at least once."
[MoneyToday Startup Media Platform Unicorn Factory]
in FUTURE
