Automated detection of circular marker particles in synchrotron phase contrast X-ray images of live mouse nasal airways for mucociliary transit assessment

• Published in: Expert systems with applications 2017-05, Vol.73, p.57-68
• Main Authors: Jung, Hye-Won, Lee, Sang-Heon, Donnelley, Martin, Parsons, David, Lee, Ivan
• Format: Article
• Language: English
• Subjects: Airway management; Algorithms; Automation; Circle detection; Circularity; Cystic fibrosis; Edge projection; Image contrast; Image detection; Imaging techniques; Medical diagnosis; Mucus; Phase contrast; Sectored ring mask; Shape recognition; Studies; Transit; X-rays
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2016.12.026

•The non-invasive measurement of mucociliary transit system for CF is required.•The automatic circular particles is challenging in Synchrotron X-ray images.•A noble method to automatically count the circular shapes is proposed.•Robust detection accuracy of 92.7% F-measurement is achieved. Cystic Fibrosis is a genetic disease in which the production of thick sticky mucus compromises the mucociliary transit (MCT) system and causes obstruction of the conducting airways. This results in a cycle of inflammation and infection that dramatically reduces quality of life and causes an early death for most. To directly assess airway health and the effects of potential treatments, synchrotron X-ray imaging techniques have been developed to non-invasively quantify MCT, by visualizing the motion of micron-sized spherical particles deposited into the nasal airways of live mice. Since the level of contrast between the target particles and the background is quite low, and the particles often overlap, most existing methods show a low detection accuracy for the MCT tracking particles in these state-of-the-art PCXI images. This paper proposes a new way to automatically detect the circular shapes of micron-sized particles in these low-contrast X-ray images. The proposed algorithm uses a gradient-directional, sectored ring mask, combined with an edge projection into the ring boundary to identify circular shapes. This new algorithm achieves significantly improved marker particle detection rate, 92.1% precision, 93.9% recall and 92.7% F-measurement, compared to existing methods. It can detect a certain degree of overlapping particles that existing methods struggle to achieve. This algorithm provides automatic MCT particle counting, which significantly reduces the manual labelling process for MCT analysis of living animals.