Interpreting Traditional Diagnostic Methods Through the Lens of Machine Learning

AI Confirms ‘Exterior–Interior’ Differentiation as the Most Crucial Standard in the First Step of Diagnosis

A research team led by Professor Kim Chang-up of the College of Korean Medicine at Gachon University has applied artificial intelligence (AI) methods to analyze pattern identification (辨證), the core diagnostic process in Korean medicine, offering a new scientific interpretation.

Pattern identification is the process of organizing a patient’s diverse symptoms into several key criteria to reach a diagnosis. The team compared this to the data analysis method of dimensionality reduction—simplifying complex symptoms using standards such as Exterior–Interior (표리, the depth of illness) or Cold–Heat (한열, temperature patterns) to make diagnosis more systematic. Their study was based on clinical cases recorded in the classic medical text Shanghanlun (Treatise on Cold Damage).

The analysis found that distinguishing Exterior–Interior was the most crucial initial step. This classification alone effectively linked complex symptoms with medicinal prescriptions, playing a central role in making subsequent diagnosis and treatment more logical and efficient. This aligns with the traditional view of Qing dynasty physician Zheng Guopeng, who emphasized “examining Exterior–Interior first” in the diagnostic sequence.

Experiments using an AI decision-tree model confirmed the same result. When trained to select medicinal herbs based solely on symptom data, the model’s very first criterion was always the Exterior–Interior distinction. Moreover, when this concept was included in the dataset, prescription prediction accuracy improved significantly. The research team explained that this opens a pathway to scientifically analyzing what has long been considered the subjective diagnostic process in Korean medicine.

Professor Kim Chang-up commented, “This study establishes a foundation for quantitatively and objectively analyzing the clinical reasoning process of Korean medicine. Based on these results, we aim to systematize diagnostic procedures and develop AI-assisted systems for use in both education and clinical practice.”

The study was led by Professor Kim, with contributions from Dr. Bae Hyo-jin (co–first author, now Professor at Dongguk University College of Korean Medicine) and doctoral researcher Kang Bong-su of Gachon University. The findings were published in the October 2025 issue of Computers in Biology and Medicine (Impact Factor 6.3, top 5.2% in JCR for mathematics and computational biology). The research was supported by the Ministry of Science and ICT and the Ministry of Education of Korea.