K-MIMIC: a nationwide Korean multi-institutional Multimodal intensive care dataset

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Advancements in critical care necessitate comprehensive, multimodal datasets for clinical decision-making and AI research.
• Existing multimodal datasets are scarce, particularly in Asian healthcare systems.
• The Korean Multi-Institutional Multimodal Intensive Care (K-MIMIC) dataset was developed to address this gap.

Purpose of the Study:

• To create the first large-scale, multicenter, multimodal intensive care unit (ICU) dataset in Asia.
• To facilitate clinical decision-making and advance artificial intelligence (AI) research in critical care.
• To enable temporal trajectory analysis through multimodal data linkage.

Main Methods:

• A retrospective multicenter study collected ICU data from 10 hospitals in Korea (2001-2023).
• Data included structured electronic medical records (EMRs), physiological waveforms (bio-signals), and medical imaging.
• Standardized protocols and de-identification procedures were followed, ensuring compliance with Korean Health Data Utilization Guidelines.

Main Results:

• The K-MIMIC dataset comprises 287,274 ICU admissions from 241,805 patients.
• Includes 22,588 bio-signal files and 496,999 imaging studies, with EMRs linked.
• 15,548 patients had data across all three modalities (EMR, bio-signals, imaging) for full multimodal analysis.

Conclusions:

• K-MIMIC is the first large-scale, multicenter, multimodal ICU dataset in Asia.
• It serves as a robust resource for critical care research, including AI-based prediction and monitoring.
• Demonstrates the feasibility of secure, standardized ICU data integration across diverse institutions.