Back to Journal

SM Journal of Public Health & Epidemiology

A Computerized Algorithm of Extracting Acoustic Measures from Prenatal Diagnostic Ultrasound Images

[ ISSN : 2473-0661 ]

Abstract
Details

Received: 15-Mar-2018

Accepted: 19-Apr-2018

Published: 20-Apr-2018

Fagen Xie¹*, Chaba Sandor², Don A Lewis² and Darios Getahun¹

¹Department of Research and Evaluation, Kaiser Permanente Southern California, USA
²Department of Imaging Solutions and Strategy, Kaiser Permanente Southern California, USA

Corresponding Author:

Fagen Xie, Department of Research and Evaluation, Kaiser Permanente Southern California Medical Group, 100 S. Los Robles Ave, 2nd Floor, Pasadena CA 91101, USA, Tel: (626) 564-3294; Fax: (626) 564-7872; Email: fagen.xie@kp.org

Keywords

Diagnostic ultrasound image; Mechanical index; Thermal index; Transducer frequency; Optical character recognition; Matching pattern; Computerized algorithm

Abstract

Background: Diagnostic Ultrasound (dUS) images capture acoustic measures. Higher level of acoustic measures may increase the likelihood of adverse neurodevelopment.

Objective: To develop a computerized algorithm to extract acoustic measures from dUS images.

Method: The dUS images of 484 pregnant women in 2014 were extracted from the Electronic Medical Record (EMR) system within an integrated healthcare organization. The retrieved dUS images were processed by the optical character recognition engine, Tesseract, to recognize the embedded texts. A set of matching patterns was constructed to extract the values associated with Thermal Index (TI), Mechanical Index (MI) and transducer frequency from these recognized texts. A sample of 200 randomly selected dUS images was processed by the computerized algorithm and results were compared against the gold standard of perinatal expert reviews.

Results: 54,909 dUS images were extracted from the EMR system. 52,637 of them had at least one of acoustic measures. The mean of extracted TI, MI and transducer frequency were 1.05, 1.08 and 4.34(MHz), respectively. Higher frequencies of dUS (5-7 MHz), higher MI (≥1.00) and higher TI (≥1.00) were used during first trimester, first/second trimesters and second/third trimesters, respectively. The computerized algorithm achieved a performance with sensitivity of 99.0%, 93.3%, 62.0% and positive predictive value of 100.0%, 99.5%, 95.8% for TI, MI and transducer frequency, respectively.

Conclusions: Our study successfully developed a computerized algorithm to extract TI, MI and transducer frequency from dUS images. Implementation of this algorithm can provide values for examining potential effects of acoustic measures on perinatal outcomes and evidence-based decision making.

Citation

Xie F, Sandor C, Lewis DA and Getahun D. A Computerized Algorithm of Extracting Acoustic Measures from Prenatal Diagnostic Ultrasound Images. SM J Public Health Epidemiol. 2018; 4(1): 1046s1.