Estimating the deposition efficiency of micro-particles in human upper airway using computerized tomography imaging and bio-inspired evolvable extreme learning machine interpolator
Affiliations: [a] Department of Mechanical Engineering, Babol University of Technology, Babol, Iran | [b] Systems Design Engineering Department, University of Waterloo, ON, Canada | [c] Department of Biomedical Engineering, Amirkabir University of Technology, Tehran, Iran
Correspondence:
[*]
Corresponding author: Ahmad Mozaffari, Systems Design Engineering Department, University of Waterloo, Ontario, Canada. E-mail:[email protected]
Abstract: In this investigation, an intelligent technique is used to analyze the
deposition efficiency (DE) of micro-particles in a realistic human upper
airway model. To do so, firstly a numerical framework including computerized
tomography (CT) imaging, geometry production software and mesh generation
tool is utilized to provide a bed for simulating the human airway.
Thereafter, the simulated airway is exposed to computational fluid-particle
dynamics (CFPD) system to study the complete upper trachea-bronchial airway
from trachea (G0) to second generation of bifurcations (G2). At the
numerical phase, low Reynolds number (LRN) k-ω turbulence model is
considered to simulate the laminar to turbulent occurred airflow. At the
final stage, the obtained knowledge is used to provide a continuous model
suited for analyzing the DE value in different sections of human upper
airway. Here, the authors utilize a fast yet accurate intelligent
interpolator called extreme learning machine (ELM) neural network to capture
the knowledge of the obtained database. Bio-inspired metaheuristics are also
used to evolve the architecture of ELM neural network such that it can
predict the DE values with a high degree of accuracy and robustness. The
results of the conducted experiments indicate that the proposed bio-inspired
ELM interpolator is capable of providing a fast and accurate model which can
yield proper information in a very short period of time as compared to the
existing numerical techniques such as CFD. On the other hand, by using
different types of bio-inspired metaheuristics, i.e. mutable smart bee
algorithm (MSBA), particle swarm optimization (PSO), firefly algorithm (FA)
and scale-factor local search differential evolution (SFLSDE), it is
observed that the structure of ELM neural network can be easily trained
using metaheuristics. In general, the experimental results demonstrate the
applicability and efficacy of soft computing techniques for predicting the
DE values in human upper airway.
