D.S. Shchepetov

The paper consists of three parts 1) An introduction 2) The main part 3) The results of algorithm testing The main difficulties of automatic P-wave detection on the continuous ECG records are listed in the "introduction" section. The principle ones are: small amplitude of P-wave, variability of P-wave amplitude and shape, overlay between the P-wave considered and previous PQRST complex T-wave e.t.c. The "algorithm description" section could be divided into two parts. The preface contains the definition of the notion - P-wave area of search - and the enumeration of the main assumptions about the structure of the considered signal ECG(t). The main part describes the solution of P-wave search problem. To simplify the description of algorithm we will solve three simplified problems of P-wave search instead of solving the general one. The first problem is searching of P-wave under the assumption that there is only one P-wave in the P-wave area of search, and its shape and width are given. To solve this problem we use special filter to select most P-wave like parts of ECG. The form of optimal filter is introduced as a function of the shape and width of the P-wave. The second problem is more complicated that the first one. We still have only one P-wave in the area of search, but its parameters are unknown. To solve this problem we replace the filtration of the one optimal filter with the filtration of the bank of filters with the selection of the best one. We introduce an approach that allows to find the best filter and position for O(N*M), where N denotes the length of the area of search; M denotes the length of the filter. Thus we could process the entire holter record very fast. The third one is closest to the real case. We have no information about number of P-waves in the area of search, or about it parameters. For solving this problem we introduce the approach that allow to select all non-intersecting P-wave like parts of ECG and to test it to be close to the real P-wave by electrophysiological parameters. Taking into account the number of found P-waves we deduce the kind of P-wave search section of ECG: the absence of P-wave, atrial fibrillation, the presence of P-wave. In the "results" section the results of the algorithm testing on synthetic signals, AHA-DB and MIT-BIH bases are presented. Also one can find measures of algorithm performance here.