Development of an innovative technology to segment luminal borders of intravascular ultrasound image sequences in a fully automated manner. (May 2019)

• Record Type: Journal Article
• Authors: Moshfegh, Abouzar; Javadzadegan, Ashkan; Mohammadi, Maryam; Ravipudi, Lakshitha; Cheng, Shaokoon; Martins, Ralph
• Is Part Of: Computers in biology and medicine. Volume 108(2019)
• Page Start: 111
• Page End: 121
• Publication Date: 2019-05
• Subjects: intravascular imaging -- Real-time lumen detection -- Machine learning -- Fully automated analysis; Medicine -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; 610.285
• DOI: 10.1016/j.compbiomed.2019.03.008
• Languages: English
• ISSNs: 0010-4825

Abstract

Abstract: Although intravascular ultrasound (IVUS) is the commonest intravascular imaging modality, it still is inefficient for clinical use as it requires laborious manual analysis. This study demonstrates the feasibility of a near real-time fully automated technology for accurate identification, detection, and quantification of luminal borders in intravascular images. This technology uses a combination of the novel approaches of a self-tuning engine, dynamic and static masking systems, radar-wise scan, and contour correction cycle method. The performance of the computer algorithm developed based on this technology was tested on a sequence of IVUS and True Vessel Characterization (TVC) images obtained from the left anterior descending (LAD) artery of 6 patients with coronary artery disease. The accuracy of the algorithm was evaluated by comparing luminal borders traced manually with those detected automatically. The processing time of the developed algorithm was also tested on a Dell laptop with an Intel Core i7-8750H Processor (4.1 GHz with 6 cores, 9 MB Cache). Linear regression and Bland-Altman analyses indicated high correlation between manual and automatic tracings (Y = 0.80 × X+1.70, R 2 = 0.88 & 0.67 ± 1.31 (bias±SD)). Whereas analysis of 2000 IVUS images using one CPU core with a 30% load took 23.12 min, the same analysis using six CPU cores with 90% load took 1.0 min. The performance, accuracy, and speed of the presented state-of-the-art technology demonstrates its capacity for use in clinical settings. Highlights: Real-time and fully automated segmentation of intravascular images is feasible. Use of longitudinal sequence as an aid in accurate segmentation is not necessary. Smart and dynamic masking systems are essential to attenuate artifacts for accuracy. Reasonable statistical accuracy was obtained by comparison with clinical contours. The accuracy and performance of the developed method is suitable for clinical use.