A research team led by Professor Jinsung Park of the Department of Biomechatronics Engineering(co–first authors: Eugene Park, M.S.; Dr. Hyunjun Park; Dr. Woochang Kim) has developed the world’s first AI-based optical diagnostic platform through a collaborative study with Dr. Minhee Kang of the Biomedical Engineering Research Center at Samsung Medical Center and Professor Gwanghui Ryu’s Otolaryngology team. This platform enables rapid and accurate differentiation—within minutes—between ordinary nasal secretion and cerebrospinal fluid (CSF) leaking from the nose.

Cerebrospinal fluid (CSF) is a vital liquid that circulates around the brain and spinal cord, protecting them from external shocks. However, due to head trauma, aging, or transnasal brain surgery, CSF can leak through the nasal cavity—a condition known as CSF rhinorrhea. Because leaked CSF appears as a clear, water-like fluid, it is visually indistinguishable from normal nasal secretion. As a result, many patients mistakenly attribute the symptom to rhinitis or a common cold and delay treatment, allowing bacteria to enter the brain and potentially cause life-threatening complications such as meningitis.

To address this challenge, Professor Park’s team focused on Raman spectroscopy, an analytical technique that reads the molecular “fingerprints” of substances through light scattering. The researchers fabricated nanoscale pillar structures composed of gold and silver, dramatically amplifying the weak signals of various biomolecules in liquid samples by tens of thousands of times. By integrating artificial intelligence (AI)–based machine learning, the system was trained to autonomously learn and distinguish the distinct spectral patterns of CSF and nasal secretions.

When evaluated using clinical samples from patients at Samsung Medical Center, the platform achieved an exceptionally high diagnostic accuracy of 90.8% in identifying CSF leakage. Notably, the researchers introduced a specialized calibration algorithm to overcome variations in spectral resolution across different Raman instruments. As a result, the platform delivered equally accurate performance not only on high-end hospital equipment but also on compact, portable devices. This advancement suggests the potential for near-instant diagnosis within approximately one minute even in emergency rooms or small outpatient clinics.

By presenting the world’s first AI-based optical diagnostic platform capable of distinguishing visually indistinguishable nasal secretion and CSF, this study overcomes a long-standing limitation in the immediate clinical confirmation of CSF leakage. The proposed technology is expected to serve as a reliable monitoring and diagnostic platform for patients suspected of CSF rhinorrhea in real-world medical settings.

This research was supported by the National Research Foundation of Korea through the Mid-Career Research Program (No. NRF-2023R1A2C2004964), the Bio & Medical Technology Development R&D Program (RS-2024-00438542), and the Sejong Science Fellowship (RS-2025-00554830, RS-2024-00353529), as well as the SKKU–SMC Future Convergence Research Program and the SKKU–KBSMC Future Clinical Convergence Research Program. In recognition of its scientific excellence, the study was published online on December 3 in the Journal of Materials Science & Technology (Impact Factor: 14.3), one of the world’s leading journals in metallurgy and materials science.

• ※Title: Ultrasensitive CSF rhinorrhea screening via machine learning-aided SERS on Au@Ag nanopillars
• ※Journal: Journal of Materials Science & Technology
• ※DOI: https://www.sciencedirect.com/science/article/pii/S1005030225012083

※PURE: https://pure.skku.edu/en/persons/jinsung-park-2/

Schematic illustration of the development of the AI-based CSF leakage diagnostic platform

Morphology and SERS characteristics of the optical substrate, the core component of the platform

Raman spectroscopic SERS detection results of cerebrospinal fluid (CSF) and nasal secretion (NS) samples

Comparison, validation, and interpretation of prediction results across various machine-learning pipelines

Application of the cross-instrument spectral preprocessing (CISP) algorithm to overcome inter-instrument resolution differences and platform validation using a portable Raman spectrometer