Detection of peripheral arterial disease using Doppler spectrogram based expert system for Point-of-Care applications. (September 2019)

• Record Type: Journal Article
• Authors: Jana, Biswabandhu; Oswal, Kamal; Mitra, Sankar; Saha, Goutam; Banerjee, Swapna
• Is Part Of: Biomedical signal processing and control. Volume 54(2019)
• Publication Date: 2019-09
• Subjects: Ultrasonography -- Peripheral artery disease -- Features extraction -- Machine learning -- Android; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.101599
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A low-cost peripheral arterial disease detection system based on the Doppler blood flow spectrogram of lower limb arteries is proposed. Hemodynamic features of the spectrogram are employed in a machine learning framework for detecting inconsistent blood flow, locating the zone of arterial stenosis and grading the stenosis. The system is implemented on a smartphone for point-of-care applications. Abstract: Peripheral arterial disease (PAD) is a common manifestation of cardiovascular diseases and more prevalent in underdeveloped countries. Ultrasound (US) is one of the preferred non-invasive diagnostic techniques for the evaluation of PAD. This work aims at achieving a low-cost PAD detection technique for mass screening. A computer aided diagnosis (CAD) method has been proposed based on the Doppler blood flow spectrograms of lower limb arteries. The proposed scheme initially removes noise from the spectrogram (350 × 175 pixels) and extracts the hemodynamic features which are generally independent of the Doppler angle. From these, best feature subsets are selected using the wrapper algorithm and supervised classifiers are developed in a machine learning framework to perform using 10-fold cross-validation technique. Overall, 334 arterial segments of 60 subjects are investigated where reference measurement is taken from the triplex mode US scanning. The quantitative assessment using random forest based classifier provides an accuracy of 84.37% and 87.93% for detecting the blood flow irregularities in above-knee and below-knee arterial segments, respectively. To classify the arterial diseases into normal, stenosis and occlusion categories, support vector machine (SVM) classifier is found to provide 97.91% accuracy on the unknown testing dataset. Moreover, variations of diagnostic parameters around the proximal and distal arterial segments define the zone of significant stenosis. The degree of stenosis is determined to quantify the severity of obstruction and the accuracy for stenosis greater than 50% is found to be 96.83%. Finally, smartphone application is implemented to provide a cost-effective, portable, user-friendly solution for Point-of-Care US system.