Tracking of multiple cells with ant pheromone field evolution. (June 2018)

• Record Type: Journal Article
• Authors: Lu, Mingli; Xu, Benlian; Nener, Brett
• Is Part Of: Engineering applications of artificial intelligence. Volume 72(2018)
• Page Start: 150
• Page End: 161
• Publication Date: 2018-06
• Subjects: Ant colony optimization -- Cell tracking -- Parameter estimation; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2018.03.015
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Tracking of biological cells is necessary to improve the understanding of their growth and behavior. Most methods used in cell tracking are time consuming and inaccurate for large population density or closely interacting cells. In this paper, a fast and accurate ant-inspired estimating algorithm for tracking multiple cells is proposed that uses a dual prediction mechanism and a pheromone updating strategy. The dual prediction mechanism is novel and uses ant individual state prediction for a given colony as well as the corresponding pheromone field prediction from the previous frame to the current frame. Ant state prediction aims to guide ant clustering around cells of interest, while pheromone field prediction helps to accelerate the pheromone formation of the current frame using the Gaussian mixture model (GMM) approximation technique. To handle the problem of tracking closely interacting cells, we design a novel ant decision-making model based on the pheromone gradient information and heuristic function with two forms. The pheromone updating strategy is also a novel pheromone diffusion and deposit model to obtain the expected pheromone field for extracting cell states in collision and cohesion. We provide quantitative validation of the method using two challenging datasets characterized by cohesion and collision by comparing them with the results from recently reported approaches. Highlights: A fast and accurate ant-inspired estimating algorithm is first proposed for cohesion or collision cells tracking. A dual prediction mechanism is introduced and the corresponding models are defined. Ant decision-making mode using pheromone gradient information and the deposit model using the pheromone divergence are developed.