Improved Tactile Stimulus Reconstruction in Electrical Impedance Tomography Using the Discrete Cosine Transform and Machine Learning

Electrical impedance tomography (EIT) is a noninvasive imaging method with promising applications in skin-like tactile sensors. Image reconstruction in EIT is challenging due to its nonlinear and ill-posed nature, resulting in fuzzy edges, ringing, and boundary artifacts. These distortions reduce the accuracy of localizing and discriminating between touch stimuli. This work focuses on EIT difference imaging and proposes a novel representation scheme for changes in the conductivity distribution using the discrete cosine transform (DCT). The superior energy compaction and decorrelation properties of DCT are utilized to efficiently represent the conductivity distribution, which is decoded by a committee of local experts, machine-learning (ML)-based, inverse solvers. The performance of the method is evaluated based on prediction accuracy and reconstruction quality using the mean squared error and structural similarity index (SSIM). The DCT framework outperforms state-of-the-art works and improves SSIM up to 0.67, which is a 21.6% improvement from benchmarked methods and leads to better localization and discrimination of multiple objects.