Optimisation of electrical Impedance tomography image reconstruction error using heuristic algorithms

Abstract

Preventing living tissues’ direct exposure to ionising radiation has resulted in tremendous growth in medical imaging and e-health, enhancing intensive care of perilous patients and helping to improve quality of life. Moreover, the practice of image-reconstruction instruments that utilise ionising radiation significantly impacts the patient’s health. Prolonged or frequent exposure to ionising radiation is linked to several illnesses like cancer. These factors urged the advancement of non-invasive approaches, for instance, Electrical Impedance Tomography (EIT), a portable, non-invasive, low-cost, and safe imaging method. EIT image reconstruction still demands more exploitation, as it is an inverse and ill-conditioned problem. Numerous numerical techniques are used to answer this problem without producing anatomically unpredictable outcomes. Evolutionary Computational techniques can substitute conventional methods that usually create low-resolution blurry images. EIT reconstruction techniques optimise the relative error of reconstruction using population-based optimisation methods presented in this work. Three advanced optimisation methods have been developed to facilitate the iterative procedure for avoiding anatomically erratic solutions. Three different optimising techniques, namely, (a) Advanced Particle Swarm Optimisation Algorithm, (b) Advanced Gravitational Search Algorithm, and (c) Hybrid Gravitational Search Particle Swarm Optimization Algorithm (HGSPSO), are used. By utilising the advantages of these proposed techniques, the convergence and solution stability performance is improved. EIT images were obtained from the EIDORS library database for two case studies. The image reconstruction was optimised using the three proposed algorithms. EIDORS library was used for generating and solving forward and reverse problems. Two case studies were undertaken, i.e. circular tank simulation and gastric emptying. Thus, the results are analysed and presented as a real-world application of population-based optimisation methods. Results obtained from the proposed methods are quantitatively assessed with ground truth images using the relative mean squared error, confirming that a low error value is reached in the results. The HGSPSO algorithm performs better than the other proposed methods regarding solution quality and stability.