Computer based pattern recognition of carotid artery Doppler signals for disease classification: prospective validation

Signs of carotid calcification from dental panoramic radiographs are in agreement with Doppler sonography results

BACKGROUND

Duplex ultrasonography (DS) is a frequently used noninvasive method for assessing carotid artery stenosis. The level of agreement between panoramic radiographs (PMX) findings of radiopacities in the area of C3-C4 and DS results has not been established.

AIMS

(1) to examine the level of agreement between DS results and PMX signs of carotid calcification and (2) to evaluate the association between periodontitis and DS results.

MATERIAL AND METHODS

Eighty-three subjects who had received a DS assessment at the University of Washington Medical Center within 36 months volunteered for a periodontal examination, including assessments of probing pocket depth (PPD), attachment level (PAL), evidence of bleeding on probing and bone loss from PMX. Two examiners independently analyzed the radiographs for evidence of carotid calcifications. The distance between the cemento-enamel junction (CEJ) to bone level (BL) CEJ-BL was used to assess alveolar bone loss as a criteria for periodontitis.

RESULTS

Twenty-nine subjects (34.9%) presented with positive DS readings. The Mantel-Haentszel common odds ratio estimate for a positive DS score and periodontitis (> 30% of teeth with distance CEJ-BL > or = 4.0 mm) was 38.4 (95% CI: 10.6-138.7, p < 0.0001). For nonsmokers only (n = 72) the odds ratio was 43.0 (95% CI: 16.7-1178.0, p < 0.0001). Evidence of bleeding on probing was 16% of sites both in the DS-positive and -negative subjects. Subjects with a positive DS result had significantly more teeth with clinical evidence of attachment loss > or = 5.0 mm (p < 0.001). The odds ratio of having periodontitis (CEJ-BL > or = 4.0 mm at > or = 30% of the teeth) and medical records confirmed diagnosis of either a stroke or an infarct or both was 7.8 (95% CI: 2.6-23.8, p < 0.001).

CONCLUSIONS

Subjects with positive DS readings of the carotid arteries due to calcified arterial plaque are accurately detected by means of conventional PMX. The likelihood of being DS positive and having radiographic evidence of periodontitis is high. A dose-response relationship between the extent of carotid calcification and severity of periodontitis was demonstrated, supporting the hypothesis of an association between periodontitis and cardiovascular diseases.

Characterization of common carotid artery blood-flow waveforms in normal human subjects

Knowledge of human blood-flow waveforms is required for in vitro investigations and numerical modelling. Parameters of interest include: velocity and flow waveform shapes, inter- and intra-subject variability and frequency content. We characterized the blood-velocity waveforms in the left and right common carotid arteries (CCAs) of 17 normal volunteers (24 to 34 years), analysing 3560 cardiac cycles in total. Instantaneous peak-velocity (Vpeak) measurements were obtained using pulsed-Doppler ultrasound with simultaneous collection of ECG data. An archetypal Vpeak waveform was created using velocity and timing parameters at waveform feature points. We report the following timing (post-R-wave) and peak-velocity parameters: cardiac interbeat interval (T(RR)) = 0.917 s (intra-subject standard deviation = +/- 0.045 s); cycle-averaged peak-velocity (V(CYC)) = 38.8 cm s(-1) (+/-1.5 cm s(-1)); maximum systolic Vpeak = 108.2 cm s(-1) (+/-3.8 cm s(-1)) at 0.152 s (+/-0.008 s); dicrotic notch Vpeak = 19.4 cm s(-1) (+/-2.9 cm s(-1)) at 0.398 s (+/-0.007 s). Frequency components below 12 Hz constituted 95% of the amplitude spectrum. Flow waveforms were computed from Vpeak by analytical solution of Womersley flow conditions (derived mean flow = 6.0 ml s(-1)). We propose that realistic, pseudo-random flow waveform sequences can be generated for experimental studies by varying, from cycle to cycle, only T(RR) and V(CYC) of a single archetypal waveform.

A computer-based statistical pattern recognition for Doppler spectral waveforms of intracranial blood flow

A computer-based statistical pattern recognition system has been developed for the analysis of transcranial Doppler (TCD) spectral waveforms of the intracranial middle cerebral artery with varying degrees of increased intracranial pressure. This system extracts multidimensional features from TCD waveforms and performs a cluster analysis of those features. The system can automatically recognize the pattern of spectral waveform and classify it as a normal, abnormal, or borderline subclass of TCD spectral waveform. An optimum decision function was generated based on the Bayes Gaussian classifier. The accuracy of the Bayes Gaussian model the spectral waveforms reaches 100% by estimating posterior probability and using the resubstituting method of estimating misclassification in the training TCD data.

Beneficial effects of colestipol-niacin therapy on the common carotid artery. Two- and four-year reduction of intima-media thickness measured by ultrasound

BACKGROUND

Controlled clinical trials have reported treatment effects evaluated with serial imaging in coronary and femoral but not cervical arteries. The Cholesterol Lowering Atherosclerosis Study, a coronary, cervical, and femoral angiographic trial of colestipol plus niacin, included a pilot study of standardized carotid ultrasound imaging.

METHODS AND RESULTS

Seventy-eight subjects had ultrasound studies at baseline, 2, and 4 years. Twenty-four drug and 22 placebo subjects had carotid ultrasound images at baseline, 2, and 4 years with matching cervical angiograms. Computer image processing was applied to ultrasound images of common carotid (far wall) and cervical angiograms. Computer operators were blind to treatment group. Carotid ultrasound measurements were tested for treatment effects and compared with measurements of atherosclerosis in coronary and cervical angiograms. Drug subjects showed significant progressive reduction in carotid thickness at 2 (P = .0001) and 4 years (P = .0001); placebo subjects significantly increased wall thickness at 2 and 4 years. Reduced levels of apolipoprotein B and increased levels of high density lipoprotein cholesterol and apolipoprotein C-III were significant predictors of carotid wall thinning. Ultrasound-measured carotid intima-media thickness was correlated at baseline with visually read coronary angiographic stenosis and at 2 years with a robust computer measurement of mild carotid atherosclerosis.

CONCLUSIONS

Common carotid intima-media thickening can be reduced by colestipol-niacin treatment. Two-year image-processed carotid ultrasound trials can provide adequate power with 50 subjects per group to test for this treatment effect.

Effect of ensemble averaging on amplitude and feature variabilities of Doppler spectrograms recorded in the lower limb arteries

The objective of the present study was to analyse the effect of averaging Doppler blood flow signals in lower limb arteries on amplitude and feature variabilities. Doppler signals recorded in 41 iliac and 35 superficial femoral arteries having different categories of stenosis were averaged over 1-15 cardiac cycles. Based on the relative decreasing rate of an index of variability, results indicated that amplitude variability of the spectrograms was exponentially reduced to 30, 6 and 1 per cent when averaging five, ten and 15 cardiac cycles, respectively. Nine diagnostic features were extracted from Doppler spectrograms and their variations from one cardiac cycle to the next quantified. Based on the relative decreasing rate of these variations, results indicated that feature variability was exponentially reduced to 30, 4 and 1 per cent when averaging five, ten and 15 cardiac cycles, respectively. The effect of averaging on the discriminant power of the features to separate the different categories of stenosis was also investigated by performing t-test analyses. Results showed that averaging between five and ten cardiac cycles provided the better discriminant power for most cases, whereas averaging over more than ten cardiac cycles was of little benefit. Based on the spectral analysis technique used in the present study, we conclude that averaging over ten cardiac cycles is sufficient for the analysis of Doppler spectrograms recorded in the lower limbs.

The effect of averaging cardiac Doppler spectrograms on the reduction of their amplitude variability

The effect of averaging cardiac Doppler spectrograms on the reduction of their amplitude variability was investigated in 30 patients. Beat-to-beat variations in the amplitude of Doppler spectrograms were also analysed. The quantification of amplitude variability was based on the computation of the area under the absolute value of the derivative function of each spectrum composing mean spectrograms. Fast Fourier transform using a Hanning window was used to compute Doppler spectra. Results obtained over systolic and diastolic periods showed that the reduction of amplitude variability followed an exponentially decreasing curve characterised by the equation f (r) = 100 e-beta(r-1), where r is the number of cardiac cycles, beta the exponentially decreasing rate, and 100 the normalised variability for r = 1. In systole, the decreasing rate beta was 0.165, whereas in diastole it was 0.225. Reductions of the variability in systole for a number of cardiac cycles of 5, 10, 15, and 20 were 48, 77, 90 and 96 per cent, respectively. In diastole, reductions of the variability for the same numbers of cardiac cycles were 59, 87, 96 and 99 per cent, respectively. Based on these results, it can be concluded that no significant improvement in the reduction of amplitude variability may be obtained by averaging more than 20 cardiac cycles.

A Doppler signal simulator

A simulated Doppler signal generator has been developed using computer controlled band-pass filtering of a broad-band noise source. The instrument may be used to generate simulated Doppler signals of known spectral characteristics in order to test Doppler signal analysers.

Measurement of arterial blood flow by Doppler ultrasound

A review is given of quantitative techniques and clinical applications of arterial Doppler ultrasound. The currently available Doppler equipment of stand-alone continuous wave and pulsed wave units, duplex systems and colour flow systems is briefly described. Doppler ultrasound can be divided into procedures concerned with waveform analysis, volume flow measurement and more recently colour flow imaging. Arterial Doppler waveform analysis is considered for a number of areas including carotid, lower limb, renal and renal transplant, obstetrics, adult cerebral, neonatal cerebral, and tumour studies. Using a duplex scanner volume flow in arteries can be measured from estimates of vessel cross sectional area, mean Doppler frequency and beam-vessel angle. The errors associated with each of these measurements is discussed, and reports of experimentally determined in vivo accuracy of volume flow measurements made using this technique are considered. Other volume flow measurement techniques including the promising attenuation compensation method are also explored.

Real-time system for robust spectral parameter estimation in Doppler signal analysis

In assessing the level of stenosis in extracranial Doppler analysis, spectral analysis has until now been used qualitatively, for the most part. Owing to the many variables affecting the measurements (mainly noise level and instrument setting made subjectively by the operator), the reliability of the inferences on the degree of stenosis is not clearly definable. Under such conditions the need arises for algorithms and systems that can estimate spectral parameters with a higher degree of accuracy, to verify whether reliable inferences can indeed by made or if this technique is only a qualitative one. In the paper a real-time spectral analysis system is described. The system relies on a new spectral estimation algorithm which gives estimates with good robustness with respect to noise. Moreover, a clear measurement procedure which eliminates the many subjective factors affecting the estimates has also been proposed and used. The system has been evaluated with simulated signals and in clinical trials and has shown better performance than the commonly used commercial analysers.

Automatic detection of sounds and murmurs in patients with Ionescu-Shiley aortic bioprostheses

The problems encountered in the automatic detection of cardiac sounds and murmurs are numerous. The phonocardiogram (PCG) is a complex signal produced by deterministic events such as the opening and closing of the heart valves, and by random phenomena such as blood-flow turbulence. In addition, background noise and the dependence of the PCG on the recording sites render automatic detection a difficult task. In the paper we present an iterative automatic detection algorithm based on the a priori knowledge of spectral and temporal characteristics of the first and second heart sounds, the valve opening clicks, and the systolic and diastolic murmurs. The algorithm uses estimates of the PCG envelope and noise level to identify iteratively the position and duration of the significant acoustic events contained in the PCG. The results indicate that it is particularly effective in detecting the second heart sound and the aortic component of the second heart sound in patients with Ionescu-Shiley aortic valve bioprostheses. It has also some potential for the detection of the first heart sound, the systolic murmur and the diastolic murmur.

Asymmetry of Doppler spectrum in stenosis differentiation

The asymmetry of the spectral distribution of ultrasonic Doppler flow velocity signals, assessed using the coefficient of skewness, is discussed as a criterion of stenosis differentiation. Its performance is compared with that of the index of turbulence intensity for both in vitro and in vivo flow Doppler signals, recorded distal to a stenosis. The power spectral distributions are computed using the direct Fourier transform and maximum likelihood method. The asymmetry of spectral distribution has proved to be a more efficient criterion than the turbulence intensity. The maximum likelihood method ensures better stenosis differentiation than the direct FFT method.

A simple quantitative screening test for the detection of extracranial carotid artery disease

A simple screening test, using continuous wave Doppler ultrasound, for the detection of all grades of extracranial carotid artery disease has been described. The test is composed of two parts: (1) the determination of the direction of flow at the orbit and (2) on-line principal component factor analysis of the maximum frequency envelope of the Doppler shifted signal obtained from the common carotid artery. A total of 154 vessel segments have been investigated; 69 normal, 41 with a stenosis of 10-49%, 32 with a stenosis of 50-99% and 12 occlusions. Of these, 70 vessel segments were assessed prospectively and they formed the data base from which the principal components and the classification factor were calculated. The remaining 84 vessel segments were analysed prospectively on-line. The combined results gave an overall sensitivity to the detection of disease of 90% and a specificity of 77%. It has been shown that although classification of Doppler waveforms by principal component analysis is a fairly sophisticated technique it is possible, by careful design of the algorithms, to design a near real-time on-line analysis system for derived waveforms such as the maximum frequency envelope.