Early Diagnosis of Meningitis and Encephalitis Causes Using AI

Severance Develops AI Model
Diagnostic Accuracy Over 93%

An artificial intelligence (AI) model capable of early classification of the causes of meningitis and encephalitis has been developed.

Professor Park Yoorang from the Department of Biomedical Systems Informatics at Yonsei University College of Medicine, Lecturer Choi Bogyu, and Professor Kim Kyungmin from the Department of Neurology at Severance Hospital announced on the 27th that they have developed an AI model boasting a diagnostic accuracy of over 93% by utilizing initial data from patients with meningitis and encephalitis. The research results were published in the latest issue of the international journal eClinicalMedicine.

Meningitis and encephalitis are diseases characterized by inflammation of the central nervous system. The causes are very diverse, and symptoms and prognosis vary depending on the cause. Among these, if the cause is bacterial or tuberculous, the mortality rate is high. Even after treatment, sequelae such as cognitive dysfunction, cerebrovascular disorders, seizures, and repeated convulsions may occur, making rapid diagnosis of the cause and treatment important.

The causes of meningitis and encephalitis can be identified through culture tests, antibody tests, and others. However, certain tests take several weeks or more to yield results. In actual clinical practice, empirical treatment based on symptoms is performed until results are available, which can lead to complications.

The research team developed an AI-based classification model for the causes of meningitis and encephalitis and analyzed its effectiveness.

Based on data collected within 24 hours after admission from 283 patients with meningitis and encephalitis who visited Severance Hospital between 2006 and 2021, the team developed an AI classification model to diagnose which of four causes?autoimmune, bacterial, tuberculous, or viral?was responsible. The diagnostic factors used in the AI classification model included 77 data points such as vital signs like blood pressure and heart rate, brain CT, chest X-ray, blood tests, and cerebrospinal fluid tests.

Subsequently, the effectiveness of the developed model was validated using data from 283 patients at Severance Hospital and 220 patients with meningitis and encephalitis who visited Gangnam Severance Hospital from 2008 to 2022.

The research team analyzed the predictive performance of the AI model using five indicators including the Area Under the Receiver Operating Characteristic curve (AUROC), accuracy, and precision (positive predictive value).

AUROC, meaning the 'area under the ROC curve,' is a statistical method that indicates the diagnostic accuracy of a specific test tool for diagnosing a disease and is commonly used as a performance evaluation metric for AI models. Generally, the closer to 1, the better the performance, and models with AUROC above 0.8 are considered high-performance.

The AI model’s predictive accuracy (AUROC) was 0.94 (94%) for patients at Severance Hospital and 0.92 (92%) for those at Gangnam Severance Hospital.

To confirm the clinical applicability of the predictive model, the research team also verified accuracy on 1,197 patients with meningitis and encephalitis whose causes were unknown. The results showed that the predictions and diagnoses in actual clinical settings matched over 93% of the time.

As a result of analyzing the prediction accuracy of the AI model on patients from Severance Hospital and Gangnam Severance Hospital, the AUROC showed high predictive performance, with Severance Hospital at 0.94 (94%) and Gangnam Severance Hospital at 0.92 (92%).

Additionally, the research team separately selected 100 patients to compare the cause diagnosis results of the AI model with those of specialists from other departments and neurologists.

The AI cause classification model demonstrated a prediction accuracy of 93%, which was significantly higher compared to 34% accuracy by specialists from other departments and 75% by neurologists.

Professor Park Yoorang stated, "Through this study, we successfully built an AI model that analyzes the diverse causes of encephalitis and meningitis," adding, "In the future, by using the AI diagnostic model to predict the causes of encephalitis and meningitis in patients, it will be possible to promptly determine the appropriate treatment direction."

This content was produced with the assistance of AI translation services.