Abstract:

BACKGROUND: Current abnormal blood pressure waveform detection algorithms are developed by definition of each beat and extraction of feature at beat-level resolution. These methods are time-consuming for normal signals are much more than abnormal signals. There are few reports about detection abnormal blood flow waveforms.
OBJECTIVE: To propose a classification algorithm for detecting faulty signals in arterial blood pressure (ABP) and cerebral blood flow velocity (CBFV) which are normally caused by the calibration of the recording devices and artifacts during the data collection.
METHODS: The feature extraction process combined frequency domain and time-domain methods, which allowed a simple 2-class thresholding technique to differentiate the faulty signals from the recorded data.
RESULTS AND CONCLUSION: Marking and recognition of abnormal signals is key for pretreatment, which is decisive for accurate signal analysis. Current detection of abnormal signals is based on morphological analysis, involving all data. Calculation speed of this algorithm is fast because the analysis was based on both distinctions in frequency and time domain and not just in time-domain. For the lack of a standard of ‘abnormal’ ABP and CBFV signals and many uncertain factors in human mark, this algorithm offer a possibility of unifying the standard and avoiding the subjective influence of human. This algorithm detected the majority of abnormal signals, but some errors exist, because the identification criteria of this algorithm influence later processing. Therefore, subjective identification of some signals is unclear. Future study should establish rigorous identification criteria of abnormal signals and improve time-domain distinctions selection and identification.