Computer-Aided Detection of Respiratory Sounds in Bronchial Asthma Patients Based on Machine Learning Method.

The aim of the study is to develop a method for detection of pathological respiratory sound, caused by bronchial asthma, with the aid of machine learning techniques. Materials and Methods: To build and train neural networks, we used the records of respiratory sounds of bronchial asthma patients at different stages of the disease (n=951) aged from several months to 47 years old and healthy volunteers (n=167). The sounds were recorded with calm breathing at four points: at the oral cavity, above the trachea, on the chest (second intercostal space on the right side), and at a point on the back. Results: The method developed for computer-aided detection of respiratory sounds allows to diagnose sounds typical for bronchial asthma in 89.4% of cases with 89.3% sensitivity and 86.0% specificity regardless of sex and age of the patients, stage of the disease, and the point of sound recording.

[Method of diagnosis of the severity of the disease attack in patients with ulcerative colitis.]

To develop a mathematical model and evaluate its prognostic significance for determining the severity of attack (SA) in patients with ulcerative colitis (UC) on the basis of minimally invasive laboratory inflammatory tests and albumin. The study involved 64 patients (33 men and 31 women) with UC in the stage of active inflammation. The method for diagnosing SA was performed by determining the concentration in the blood of tumor necrosis factor alpha (TNF-α), C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), α2-globulin and albumin. Truelove and Witts criteria (1955) were used as a reference method for the evaluation of SA in patients with UC. In constructing a mathematical model, the severity of the attack of the UC emerged as a dependent variable. As independent variables or predictors - five laboratory parameters: TNF-α, CRP, ESR, α2-globulins and albumin, which had a high relationship with SA. At SA value in the range from 0 to 0.5, absence of attack or remission of the disease is determined, SA value in the range from 0.5 to 1.2 corresponds to mild severity of UC attack, SA value in the range from 1.2 to 2.2 corresponds to the moderate severity of the attack of the UC, at SA more than 2.2 one should speak of a severe attack of the UC. The model is statistically significant: the Fisher test F = 439.9; p < 0.0001, multiple R = 0.981; R2 = 0.961. The diagnostic sensitivity of this mathematical model was 98.4%, specificity - 96.7%. The index of the severity of the attack, calculated by the formula proposed by us, is able to differentiate the mild, moderate and severe SA, including the isolation of patients with a stage of remission. In this case, the method is non-traumatic, has a low cost and high sensitivity and specificity.