Systém pro stanovení a úpravu variability tepové frekvence

Abstract

This thesis focuses on developing methods to modify the multichannel fetal heart rate (fHR). These methods have the potential to reduce the number of erroneous conclusions, such as caesarean section, by physicians. Two databases, labour and pregnancy, are used in this work. The labour database, contains 12 records of multichannel abdominal electrocardiogram (aEKG) of 5 minutes duration. The pregnancy database contains 10 aEKG records with a length of 20 minutes. Both databases have a sampling rate of 500 Hz. Three methods are proposed in this paper, which are sequentially combined. Method 1, adjusts the multichannel fHR curves. Method 2, using the already given rules, applies the R-means detection adjustment. In method 3, the two previous methods are combined, with the difference of merging the R-peaks vectors, instead of the fHR vectors. For the evaluation of R-peaks, the accuracy of R-peaks detection is used, according to the parameters accuracy (ACC), sensitivity (SE), prediction accuracy (PPV), and F1. The Bland-Altman plot was used to evaluate the fHR curves, comparing its mean value μ and ±1.96σ values. All results are compared with the reference of the specified annotation. In most cases, the accuracy of the detection gave us a positive improvement of an order of magnitude compared to the unadjusted R-peaks detection. Evaluation using the Bland-Altman plot, for the labour database came out best for method 1, as it had the lowest average of the mean μ value of all records. For values of ±1.96σ, method 2 performed best. For the pregnancy database, the average of the mean μ and ±1.96σ values of all records came out best for method 3. Method 1 then came out best for the mean μ value and method 2 for the ±1.96σ value for all records from both databases. The overall best method is method 1. Its average of mean μ values for all records from both databases was almost twice as good as method 2 and method 3.