A parallel Architecture of a Genetic Algorithm for EIT Image Reconstruction

Nowadays, electrical impedance tomography (EIT) has become an interesting bio-medical imaging technique, it is non-invasive and shows great potential in medical investigation. The EIT presents high performances in conductivity mapping of the inner structure of a specific target. A set of electrodes are placed around the measured target where an excitation current is injected in a pre-defined frequency. This excitation will cause a variation of the potential distribution and consequently in the conductivity distribution of the measured medium. Therefore, boundary voltages are recorded, which will be used to provide the tomography image of the addressed cross-section. In this paper, the focus will be on the image reconstruction algorithm in order to reach high-quality images. A water tank set-up is used for the experiments where a conductive object is placed inside. Then, a genetic algorithm has been implemented in a parallel form to reduce the execution time by using a master/slave architecture. Moreover, to reach a high and accurate conductivity image, a reference technique has been integrated called the non-blind search. Further, a comparison study of genetic operators' configuration is carried out in terms of fitness values and object position identification. Nine combinations are tested, where the uniform crossover/random mutation shows the highest performance in object position identification with a fitness value equal to 0.0505. A comparison between the implemented solution and commonly used algorithms is established as well, where the genetic algorithm was able to outperform them in terms of fitness values.