Scientists from St. Petersburg University and the Almazov National Research Medical Center have developed a new method for diagnosing lung diseases and have obtained a patent for it.
This innovative method allows for highly accurate differentiation between tuberculosis and sarcoidosis through a simple examination of a venous blood sample, without the need for a biopsy. The technique relies on the mathematical analysis of immune cells using machine learning algorithms.
According to the developers, differentiating between tuberculosis and sarcoidosis remains one of the most complex challenges in modern pulmonary medicine, given the significant similarity in the symptoms of the two diseases. Experts indicate that the error rate in diagnosis can range between 40 and 60 percent, which can lead to serious consequences, as tuberculosis requires antibiotic treatment and patient isolation, while sarcoidosis is a non-contagious disease and is treated with entirely different methods.
Per-Jan Håkan Hedenmalm, director of the Laboratory of Probabilistic Methods in Analysis at the university, said: "We have devised a method that not only compares biomarkers, but also uses two mathematical inequalities developed using symbolic regression and multi-criteria optimization. After analyzing a patient's blood sample, the program identifies the most likely diagnosis based on the concentration of specific types of immune cells, with high accuracy."
To develop this model, the researchers used a range of machine learning techniques, enabling the algorithm not only to predict the diagnosis but also to identify cases where the data is insufficient to reach a definitive conclusion. In these instances, patients are automatically classified as "suspected cases" and referred for further testing, thus reducing the likelihood of misdiagnosis.
