AI Does Not Replace Doctors, But Assists Them

In 2016, Geoffrey Hinton predicted that radiologists would disappear within 10 years, but this has not actually happened. Professor Cho Taeho emphasized that AI is developing in a direction of assistance rather than replacement. The increasing workload due to an aging population and the need for rapid diagnosis are key factors in AI adoption, and the direction of AI development is focused on task assistance and efficiency.

Alzheimer's disease is irreversible once it occurs, so it is important to detect it early and administer medication before it progresses to a severe stage. The key to treatment is delaying disease progression through early detection and prescription.

AI research utilizing various data such as brain imaging, genomics, and metabolomics is underway for early diagnosis of dementia. Significant results are being derived from cross-analysis between diseases, rare variant research, and other fields.

While AI's diagnostic accuracy is already high, research has been conducted to trace back the basis of that accuracy. By analyzing which parts deep learning focused on to make its judgment, the possibility of identifying brain regions related to dementia that were not previously revealed in existing literature was discovered.

Beyond the limitations of brain imaging analysis, genomic research is being conducted to detect dementia earlier. Inspired by the language tokenization method of large language models (LLMs) like ChatGPT, attempts have been made to tokenize genomic blocks. Research is underway to discover and predict variations within LD (Linkage Disequilibrium) blocks in the genome.

Due to the vast amount of genomic information, attempts have been made to train AI using the Transformer algorithm, which is used in generative language models. The purpose of this is to efficiently review entire genomic data and obtain results that differentiate from the existing GWAS (Genome-Wide Association Study) method.

A sliding window algorithm has been developed where an AI agent analyzes the entire gene combination for 24 hours. This method allows the agent within the window to calculate not only gene mutations but also surrounding gene variations and their influence. This helps overcome the limitations of conventional methods that only checked single gene variations.

When comparing the analysis results of the existing GWAS (Genome-Wide Association Study) method and the sliding window method, the AI method more clearly visualizes key dementia-related regions. The sliding window produces distinct peaks, allowing important parts of the overall trend to be understood at a glance.

Currently, AI research focuses on elucidating the causes of dementia rather than developing early diagnosis tools. AI is being utilized in efforts to further uncover and get closer to the causes of dementia, which are not yet fully understood, and this is an important reporting stage for biological mechanisms and data research.

A Circular Sliding Window algorithm has been developed for dementia prediction research through blood metabolomic analysis. Metabolomic data has the disadvantage of being difficult to process in a window using a tabular format, so this problem was solved through a circular algorithm, like 'pass the parcel'.

To uncover the causes of dementia, research into rare variants that appear in only a few people, in addition to common variants that appear in everyone, is crucial. Existing research has focused on common variants, but rare variants may hold clues to understanding personalized dementia causes. Therefore, rare variant research is essential for understanding diverse dementia onset pathways and developing personalized treatment strategies.