| Description |
The objectives of this dissertation were: 1) to develop and evaluate neural networks for processing noninvasive oscillometric waveforms into estimates of blood pressure, 2) to construct an occlusive pad for recording oscillometric waveforms from the superficial temporal artery (STA) of the head, and 3) to evaluate the STA as a site for noninvasive estimation of blood pressure. Simultaneously recorded invasive blood pressure and noninvasive oscillometric waveforms from the forelimbs of five dogs (n = 425) were used to train and compare neural networks with conventional oscillometric algorithms. The intersubject mean differences and standard deviations of the differences between invasive and both noninvasive methods were similar (mean differences less than 5 mmHg and standard deviations of about 5 mmHg). However, average intrasubject standard deviations were consistently lower for neural networks (e.g., estimating mean blood pressure, 2.74 and 3.78 mmHg for neural networks and conventional algorithms, respectively). The neural networks were found to be less sensitive to noise and sparse data and thus provided a more robust and accurate method of estimating changes in a given subject's blood pressure. An occlusive pad for recording oscillometric waveforms from the STA was constructed and evaluated on 31 patients. Recordings were obtained from 20 of 21 thoracic intensive care unit (TICU) patients (n = 134) and 9 of 10 operating room (OR) patients (n = 94). Because of the limited amount of data available for training neural networks, oscillometric estimates were obtained with conventional arm oscillometric algorithms. TICU intersubject mean differences ± the standard deviations of the differences between invasive aortic measurements and STA oscillometric estimates of diastolic, mean, and systolic blood pressure were 4.69 ± 4.50, 14.36 ± 4.43, and 26.87 ± 7.93 mmHg, respectively. OR statistics comparing invasive radial artery measurements and noninvasive STA estimates were -2.80 ± 8.48, 4.58 ± 7.85, and 18.86 ± 8.78 mmHg, respectively. Although large, the differences between invasive aortic or radial artery measurements and noninvasive STA estimates were consistent over the range of blood pressures encountered. The standard deviations of the differences were similar to values reported in comparison studies of arm invasive and noninvasive methods. The expectation is that neural networks trained on STA oscillometric waveforms can compensate for the observed differences and thus provide reasonable estimates of central arterial blood pressure from the superficial temporal artery. |
