[ICKSH 2025] Basic principles of AI and its application in healthcare

The International Congress of the Korean Society of Hematology (ICKSH) 2025, held on March 28 at the Grand Walkerhill Seoul, featured an educational session titled “Basic Understanding and Application of AI” to highlight the latest advances in artificial intelligence, which is increasingly being applied to healthcare and research.

During the session, Professor Kim A-uk of Kangwon National University delivered a presentation titled “Artificial Intelligence Everywhere”; Professor Hwang Sang-hyun of Ulsan University College of Medicine spoke on “Revolutionizing Medical Documentation”; and Shin Soo-yong, Head of Research at Kakao Healthcare, presented “Generative AI Use Case for Medical Research.”

Professor Kim A-uk: ‘AI is Infrastructure -- healthcare is no exception’

Professor Kim A-uk of Kangwon National University gave a talk on “Artificial Intelligence Everywhere,” offering insightful perspectives on the history of AI, current technological trends, and its potential applications in healthcare.

At the beginning of his talk, Kim emphasized that AI is a universal technology that has permeated every aspect of human life—not just the exclusive domain of certain industries—and argued that it is time to regard AI not merely as a “tool” but as an “infrastructure.”

AI was conceptualized by John McCarthy and others at the Dartmouth Conference in 1956. Since then, it has gone through two “AI winters” (periods of stagnation due to unmet expectations), but rapidly advanced through milestones such as IBM’s Deep Blue defeating a chess champion in 1997, Geoffrey Hinton’s introduction of deep learning in 2006, and the breakthrough of AlexNet in image recognition in 2012.

In 2016, AlphaGo’s victory over professional Go player Lee Se-dol cemented public awareness that AI could rival human intelligence.

Kim identified three key drivers of AI’s progress: the accumulation of large-scale data, high-performance computing resources (e.g., GPUs and TPUs), and deep neural network algorithms (deep learning).

Building on this foundation, technologies like BERT (2018), multimodal models such as CLIP and DALL·E, and GPT (2022) have propelled AI into a stage where it can simulate human language and thought. Recently, development has accelerated in general AI models capable of performing multiple tasks with a single system.

In the medical field, AI is being applied in various areas such as image diagnosis, protein structure prediction (e.g., AlphaFold), pathology slide analysis, and low-dose CT/MRI reconstruction.

However, Kim also pointed to key technical challenges, including a lack of reliability, difficulties in generalization, data bias and scarcity, and issues in integrating multiple data sources.

Nevertheless, he emphasized, “AI is no longer a novelty but an essential underlying technology in medical research and clinical practice,” adding that researchers and clinicians must shift their perceptions to actively and effectively utilize AI.

Professor Hwang Sang-hyun: ‘ChatGPT and RAG are key to transforming medical Documentation’

Professor Hwang Sang-hyun of Ulsan University College of Medicine gave a lecture titled “Revolutionizing Medical Documentation,” presenting the potential of generative AI and the Retrieval-Augmented Generation (RAG) framework to transform documentation processes in healthcare.

He emphasized that Large Language Models (LLMs) are evolving from basic text generators to tools that can solve real-world problems in clinical settings. “Automation in medical documentation is not just about improving efficiency—it must also ensure accuracy and accountability,” he explained.

Hwang addressed the issue of “hallucination,” where LLMs may produce inaccurate information, and proposed RAG as a solution. RAG enhances accuracy and trust by first retrieving reliable external data (e.g., WHO guidelines) and then generating responses based on that information.

He outlined six key reasons for adopting RAG systems, particularly emphasizing the importance of incorporating up-to-date information and strengthening existing knowledge bases in the medical field.

Hwang also noted that LLMs are evolving into “agentic LLMs” capable of performing user tasks directly. For instance, the Claude model can analyze or run files from a user’s folder—an example of how LLMs could be integrated into complex clinical workflows.

Given the importance of data privacy and control in healthcare, Hwang stressed the need for local LLMs and introduced lightweight models that can run on a single GPU.

He shared various real-world use cases, including systematic review creation, article summarization, and data visualization, illustrating how generative AI is already being applied in clinical and academic environments. He concluded, “LLM is a powerful tool, but in healthcare, accuracy and source-based responses are paramount,” underscoring the need to balance tool usage with reliable information.

Kakao Healthcare unveils HRS: A generative AI-powered medical research platform

Shin Soo-yong, Head of Research and Chief Privacy Officer at Kakao Healthcare, gave a talk titled “Generative AI Use Case for Medical Research,” introducing the structure and applications of HRS (Healthcare Data Research Suite), an AI-based medical research platform.

“Nowadays, anyone with an idea can start a research project,” said Shin, highlighting how the combination of generative AI and medical data is revolutionizing the research environment.

HRS was developed to address the complexities of multicenter collaborative research, including unstructured data, lack of standardization, and privacy concerns. The platform features data curation that automatically harmonizes varying hospital code systems and language, natural language processing (NER) to extract data from unstructured texts like pathology reports, and an automated pseudonymization system using LLMs.

HRS is directly integrated with hospitals’ internal data warehouses (DWH), allowing researchers to simply input a topic or prompt, after which the LLM automatically extracts, cleans, and analyzes the data.

To address privacy concerns, HRS uses a federated learning model, enabling secure collaboration without centralizing raw data.

The research portal also features a user-friendly interface that allows researchers to carry out the entire research process with simple prompts, eliminating the need for complex coding or tools.

“LLMs are evolving beyond text generators into research partners that support the entire research process,” Shin said. “The future research ecosystem will be defined by the close integration of creative ideas and advanced technologies.”

Kakao Healthcare’s HRS is gaining attention as a next-generation medical research platform that offers both accuracy and efficiency, with growing expectations for its real-world application in clinical settings.