Computer-assisted detection of tardus parvus waveforms on Doppler ultrasound. (May 2018)

• Record Type: Journal Article
• Authors: Gunabushanam, Gowthaman; Millet, John D; Stilp, Erik; Crawford, Forrest W; McNamara, Robert L; Scoutt, Leslie M
• Is Part Of: Ultrasound. Volume 26:Number 2(2018:May)
• Page Start: 81
• Page End: 92
• Publication Date: 2018-05
• Subjects: Computer-aided diagnosis -- tardus parvus waveform -- spectral analysis -- aortic stenosis; Ultrasonic imaging -- Periodicals; Ultrasonography -- Periodicals; 616.0754305
• DOI: 10.1177/1742271X17747135
• Languages: English
• ISSNs: 1742-271X

Abstract

Objective: To determine if a novel computer-generated metric, effective acceleration time, improves accuracy for detecting tardus parvus waveforms on spectral Doppler ultrasound. Methods: Patients with echocardiography-confirmed aortic valve stenosis ( n = 132; 60 mild, 44 moderate, 28 severe) and matched controls ( n = 48) who underwent carotid Doppler ultrasound were identified through an imaging database search at a single medical center. A custom-built spectral analysis computer program generated effective acceleration time values for spectral Doppler waveforms in the carotid arteries and a receiver operating characteristic analysis was performed to determine the optimal median effective acceleration time cutoff value to detect tardus parvus waveforms. Two radiologists, blinded to subject disease status, reviewed and rated all carotid sonograms for presence of tardus parvus waveforms. Inter-rater variability was measured, and the accuracy of aortic valve stenosis detection with and without use of the effective acceleration time cutoff was calculated. Results: Receiver operating characteristic analysis revealed an optimal effective acceleration time cutoff of ≥ 48 ms with a corresponding area under the curve of 0.77 (95% CI: 0.70–0.84). Use of the effAT cutoff demonstrated an accuracy of 74%. Accuracy of visual waveform interpretation by raters ranged from 43% to 61%. Inter-rater agreement in detection of tardus parvus waveforms was 76% (136/180 cases, K = 0.44, p < 0.001). Conclusions: Detection of tardus parvus waveforms through visual interpretation of spectral Doppler waveform morphology is limited by low accuracy and moderate inter-rater variability. Use of a computer-generated median effective acceleration time cutoff value markedly improves diagnostic accuracy and avoids observer variability.