Measuring Bolus Transit Times from Videofluoroscopy using Image Profiles and Particle Swarm Optimisation

摘要

In video-fluoroscopic swallowing assessment the time taken for the bolus material to transit the oral and pharyngeal regions is currently estimated by visual inspection. This paper presents an effective method for objectively approximating these timings by fitting a Gaussian surface to the first derivative of the intensity profiles along user defined anatomical boundaries. The profile characteristics taken for every video frame generates a spatiotemporal function in which the bolus crossing can be detected. A set of heuristics based on area and eccentricity is first applied to find the correct region of interest on the spatiotemporal surface. Particle Swarm Optimisation is used to search for the best fitting Gaussian surface using the mean squared error as the cost function. The peak temporal value of the fitted surface is returned as the extracted bolus crossing time. Results show close matching when compared with visually inspected timings.