Simulated annealing implementation with shorter Markov chain length to reduce computational burden and its application to the analysis of pulmonary airway architecture

A new way to implement the Simulated Annealing (SA) algorithm was developed and tested that improves computation performance by using shorter Markov chain length (inner iterations) and repeating the entire SA process until the final function value meets the solution criterion. The new approach coupled with the adaptive neighborhood method was tested on the Rosenbrock function in 4 and 13 dimensions. This implementation significantly improved the computation speed without degrading solution quality. The proposed implementation was used to characterize pulmonary architecture from micro CT image data demonstrating the algorithm's effectiveness especially for problems with high computational demand and when the solution quality requirement can be pre-specified. Using this implementation, detailed statistics of the morphometry of conducting airways from 12 male Sprague Dawley rats were obtained for each lobe.