Electrical impedance tomography (EIT) is a low cost, radiation-free and non-invasive imaging technique. It provides high temporal resolution but relatively low spatial resolution. This paper proposes a novel method to enhance the spatial resolution of EIT images using Cycle generative adversarial networks (cycleGAN). The method consists of multiple steps including traditional image reconstruction, custom postprocessing of time series EIT images to reduce noise, CT image preprocessing to reduce unnecessary feature entanglement, EIT-to-CT transformation, and finally an anatomically constrained EIT image reconstruction. The EIT-to-CT transformation consists of a cycleGAN model trained on unpaired EIT and CT lung images in which we incorporate the Mutual Information (MI) constraint. This EIT-to-CT step provides a structurally aligned high resolution CT image inferred from the lower resolution EIT image. The subsequent EIT reconstruction uses this high resolution image as a constraint to provide physiologically relevant EIT images with enhanced resolution. The proposed method is validated with 235 EIT images acquired from over 10 healthy subjects. It is compared to the traditional image reconstruction method using the Intersection over Union (IOU) and Normalized Mutual Information (NMI) metrics, giving 0.7 versus 0.4 and 0.49 versus 0.45,p < 0.001 respectively, suggesting that the proposed method is better. While paired testing of EIT and CT images will further corroborate the results, the initial results hold promise. Finally, this pipeline of feedback oriented reconstruction based on an alignment goal inferred through cycleGAN, can potentially transfer to and aid in different medical imaging reconstruction modalities.