Yonsei researchers develop predictive AI to treat brain infections
A new model based on deep learning developed in South Korea has shown potential to reduce the time necessary to identify causes and predict the risk factors for brain and vertebral infections.
Researchers from the Yonsei University health system (YUHS) have joined a project supported by the Hyundai Motor Foundation to develop and test an AI model to diagnose and pronounce diseases of the central nervous system (SNC), such as encephalitis and meningitis.
RESULTS
They formed their AI model on nearly 1,500 3D images of cerebrospinal liquid immune cells (CSF) collected from 14 patients who presented SNC infections at Severance hospital.
The team claimed to be the first to use 3D cell images and to analyze cell morphology in the diagnosis and prediction of the prognosis of brain and vertebral infections.
In a study evaluating the AI model, the researchers found that its predictive performance increased in precision as more and more cellular images crossed it.
Initially, the execution of a cell image has shown 89% accuracy to identify the cause of the infection and a precision of 79% to predict the probable course of the disease. By nourishing five cellular images, the accuracy of the identification of causal pathogens and the prediction of the prognosis reached 99% and 94%, respectively. 100% precision in both tasks was obtained with less than 10 cellular images.
In addition, the authors have also identified the cell mass, the volume and the density of proteins as important factors to predict the prognosis.
The study Results were published in Advanced Systems Journal de Wiley.
Why it matters
The diagnosis and prognosis of diseases of the central nervous system may vary depending on the cause. Bacterial or tuberculous causes have the highest mortality rate, which, according to Yuhs researchers, critically requires rapid diagnosis and rapid treatment.
Meanwhile, there are different diagnostic tests for each causal pathogen, and some tests can take weeks or more to produce reports. The LCR test, for example, would require additional manual confirmation of the form and number of cells.
Researchers at the University of Yonsei sought to reduce this whole process – from the LCR collection to evaluation – up to an hour. “The method we offer is faster than current diagnostic techniques, such as brain imagery, and frequently used clinical biomarkers such as procalcitonin blood rates and C-reactive proteins,” they said.
They also underlined the potential of cell morphology, analyzed by 3D holotomography, as a very precise and effective “biomarker” for brain and vertebral infections. This profitable imaging technique produces 3D cell images without real -time label.
At the same time as
“This study is the first case of using three-dimensional images of immune cells in cerebrospinal fluid in order to predict the cause and prognosis of patients infections by the central nervous system. We expect the in-depth learning model presented in the study to be useful for shortening the time necessary for the diagnosis of patients and predicting prognosis,” said one of the Biomed park, Researchers and history of the department of the department of the department of the Biome Systems department.
