Automated and accurate carotid bulb detection, its verification and validation in low quality frozen frames and motion video

AIMS: cIMT measurements during clinical trials need to have a fixed reference point (also called as bulb edge points) in the anatomy from which the carotid intimamedia thickness (cIMT) can be measured. Identification of the bulb edge points in carotid ultrasound images faces the challenge to be detected automatically due to low image quality and variations in ultrasound images, motion artefacts, image acquisition protocols, position of the patient, and orientation of the linear probe with respect to bulb and ultrasound gain controls during acquisition. METHODS: This paper presents a patented comprehensive methodology for carotid bulb localization and bulb edge detection as a reference point. The method consists of estimating the lumenintima borders accurately using classification paradigm. Transition points are located automatically based on curvature characteristics. Further we verify and validate the locations of bulb edge points using combination of several local image processing methods such as (i) lumenintima shapes, (ii) bulb slopes, (iii) bulb curvature, (iv) mean lumen thickness and its variations, and (v) geometric shape fitting. RESULTS: Our database consists of 155 ultrasound bulb images taken from various ultrasound machines with varying resolutions and imaging conditions. Further we run our automated system blindly to spot out the bulbs in a mixture database of 336 images consisting of bulbs and nobulbs. We are able to detect the bulbs in the bulb database with 100% accuracy having 92% as close as to a Neurologists's bulb location. Our mean lumenintima error is 0.0133 mm with precision against the manual tracings to be 98.92%. Our bulb detection system is fast and takes on an average 9 seconds per image for detection for the bulb edge points and 4 seconds for verification/validation of the bulb edge points.