Cataracts, an eye disease that causes objects to appear blurred as though shrouded in fog, differ in lens opacity and location for each patient. In the past, doctors explained to patients by hand-drawing images on paper charts. However, as hospital systems became computerized, this also became difficult, and YIDO has addressed this. It produces images that reflect the patient's condition instantly using only the doctor's voice commands or text.
Generative AI refers to technology that creates new results in text, images, videos, music, coding, and more, by learning from big data and patterns. The representative example is the 'ChatGPT' from OpenAI, released at the end of 2022, which has recently emerged as a technology to overcome various limitations in the medical field and enhance work efficiency.
◇Generated images of patients' glaucoma conditions based on medical staff's voice
According to the medical community on the 21st, a research team led by Professor Hwang Woong-joo from the Catholic University of Korea's Yeouido St. Mary's Hospital has developed the country's first 'Ophthalmic Generative AI' based on image-generating medical software and its integrated 'Generative AI Electronic Medical Record (EMR) System.'
The Ophthalmic Generative AI developed by the research team converts voice commands from medical staff into text in real time and generates images from them. The integrated EMR system provides customized standard chart formats for retina, glaucoma, and cataracts along with the basic chart. Medical staff can use the image generation technology to create medical records in a more intuitive and convenient manner.
Professor Hwang noted, 'Current EMR systems have difficulty visualizing and recording the patient's eye condition, and the developed Ophthalmic Generative AI and EMR will help enhance intuitive understanding for patients, improve efficiency for medical staff, and reduce unnecessary manpower consumption.'
The research team also plans to develop a large language model (LLM) based AI agent EMR system. AI agents refer to technology wherein generative AI evolves to mimic human behaviors and perform tasks autonomously.
The research team has also developed the world's first glaucoma diagnostic solution using generative AI in collaboration with PuzzleAI Research Institute. It learns by matching optical coherence tomography images of glaucoma patients with fundus examination findings, and generates results for visual field tests based on this to diagnose glaucoma. Professor Hwang explained, 'This will reduce excessive redundant testing and allow for complex test result predictions through affordable examinations.'
◇Drafting interpretative reports after diagnosis... discovering and designing new drug candidates
According to industry sources, companies such as Lunit, Kakao's subsidiary KakaoBrain, and DEEPNOID are developing medical platforms incorporating generative AI technology.
Currently available medical AI software mainly assists in diagnosing diseases based on imaging but is expected to expand its scope of work to include drafting interpretative reports in the future.
For example, the generative AI platform could show what issues exist in chest imaging video instead of the doctor, and simultaneously provide a text response. If AI diagnoses pneumothorax based on the patient's X-ray, it could draft an interpretative report stating, 'Judged as pneumothorax and intubation is necessary,' along with the image, and suggest a treatment plan.
AI is also establishing itself as an important tool in drug development.
NVIDIA, based in the United States, launched the BioNEMO generative AI model for drug development earlier this year. This model has learned 'amino acid sequences and protein structures.' Based on this, when the generative AI creates a three-dimensional protein model, AI specialized in inference and argumentation derives the optimal combinations between proteins.
Syntekabio has developed a learning technique known as '3bm GPT' that utilizes the generative AI GPT model to provide analysis and prediction results regarding the binding of three-dimensional proteins and compound ligands. The company stated, 'It is possible to derive the properties of proteins or compounds even with insufficient binding information, enhancing the convenience for researchers, and it can be applied to search for substances or target proteins that can be bound.'
As various medical products utilizing generative AI technology continue to be developed both domestically and internationally, regulatory authorities are preparing related legislation. The Ministry of Food and Drug Safety plans to release Korea's first guidelines for generative AI medical devices on the 24th.
When generative AI fails to find accurate answers, it may generate incorrect responses based on similar sections among its vast learning contents, a problem known as 'hallucination.' Concerns have been raised for the need for rigorous review and approval standards for products based on generative AI, considering the risks of such errors and other various issues. The Korea Disease Control and Prevention Agency plans to establish ethical guidelines for generative AI research in the digital healthcare field by the end of the year.